diff --git a/src/uu/split/src/split.rs b/src/uu/split/src/split.rs
index 445c1f205..128ef73c6 100644
--- a/src/uu/split/src/split.rs
+++ b/src/uu/split/src/split.rs
@@ -13,11 +13,11 @@ extern crate uucore;
 mod platform;
 
 use clap::{App, Arg};
-use std::char;
 use std::env;
 use std::fs::File;
-use std::io::{stdin, stdout, BufRead, BufReader, BufWriter, Read, Write};
+use std::io::{stdin, BufRead, BufReader, BufWriter, Read, Write};
 use std::path::Path;
+use std::{char, fs::remove_file};
 
 static NAME: &str = "split";
 static VERSION: &str = env!("CARGO_PKG_VERSION");
@@ -213,50 +213,65 @@ struct Settings {
     verbose: bool,
 }
 
-struct SplitControl {
-    current_line: String,   // Don't touch
-    request_new_file: bool, // Splitter implementation requests new file
-}
-
 trait Splitter {
-    // Consume the current_line and return the consumed string
-    fn consume(&mut self, _: &mut SplitControl) -> String;
+    // Consume as much as possible from `reader` so as to saturate `writer`.
+    // Equivalent to finishing one of the part files. Returns the number of
+    // bytes that have been moved.
+    fn consume(
+        &mut self,
+        reader: &mut BufReader<Box<dyn Read>>,
+        writer: &mut BufWriter<Box<dyn Write>>,
+    ) -> usize;
 }
 
 struct LineSplitter {
-    saved_lines_to_write: usize,
-    lines_to_write: usize,
+    lines_per_split: usize,
 }
 
 impl LineSplitter {
     fn new(settings: &Settings) -> LineSplitter {
-        let n = match settings.strategy_param.parse() {
-            Ok(a) => a,
-            Err(e) => crash!(1, "invalid number of lines: {}", e),
-        };
         LineSplitter {
-            saved_lines_to_write: n,
-            lines_to_write: n,
+            lines_per_split: settings
+                .strategy_param
+                .parse()
+                .unwrap_or_else(|e| crash!(1, "invalid number of lines: {}", e)),
         }
     }
 }
 
 impl Splitter for LineSplitter {
-    fn consume(&mut self, control: &mut SplitControl) -> String {
-        self.lines_to_write -= 1;
-        if self.lines_to_write == 0 {
-            self.lines_to_write = self.saved_lines_to_write;
-            control.request_new_file = true;
+    fn consume(
+        &mut self,
+        reader: &mut BufReader<Box<dyn Read>>,
+        writer: &mut BufWriter<Box<dyn Write>>,
+    ) -> usize {
+        let mut bytes_consumed = 0usize;
+        let mut buffer = String::with_capacity(1024);
+        for _ in 0..self.lines_per_split {
+            let bytes_read = reader
+                .read_line(&mut buffer)
+                .unwrap_or_else(|_| crash!(1, "error reading bytes from input file"));
+            // If we ever read 0 bytes then we know we've hit EOF.
+            if bytes_read == 0 {
+                return bytes_consumed;
+            }
+
+            writer
+                .write_all(buffer.as_bytes())
+                .unwrap_or_else(|_| crash!(1, "error writing bytes to output file"));
+            // Empty out the String buffer since `read_line` appends instead of
+            // replaces.
+            buffer.clear();
+
+            bytes_consumed += bytes_read;
         }
-        control.current_line.clone()
+
+        bytes_consumed
     }
 }
 
 struct ByteSplitter {
-    saved_bytes_to_write: usize,
-    bytes_to_write: usize,
-    break_on_line_end: bool,
-    require_whole_line: bool,
+    bytes_per_split: usize,
 }
 
 impl ByteSplitter {
@@ -294,36 +309,44 @@ impl ByteSplitter {
         // Try to parse the actual numeral.
         let n = &settings.strategy_param[0..(settings.strategy_param.len() - suffix.len())]
             .parse::<usize>()
-            .unwrap_or_else(|_| crash!(1, "invalid number of bytes"));
+            .unwrap_or_else(|e| crash!(1, "invalid number of bytes: {}", e));
 
         ByteSplitter {
-            saved_bytes_to_write: n * multiplier,
-            bytes_to_write: n * multiplier,
-            break_on_line_end: settings.strategy == "b",
-            require_whole_line: false,
+            bytes_per_split: n * multiplier,
         }
     }
 }
 
 impl Splitter for ByteSplitter {
-    fn consume(&mut self, control: &mut SplitControl) -> String {
-        let line = control.current_line.clone();
-        let n = std::cmp::min(line.chars().count(), self.bytes_to_write);
-        if self.require_whole_line && n < line.chars().count() {
-            self.bytes_to_write = self.saved_bytes_to_write;
-            control.request_new_file = true;
-            self.require_whole_line = false;
-            return "".to_owned();
+    fn consume(
+        &mut self,
+        reader: &mut BufReader<Box<dyn Read>>,
+        writer: &mut BufWriter<Box<dyn Write>>,
+    ) -> usize {
+        // We buffer reads and writes. We proceed until `bytes_consumed` is
+        // equal to `self.bytes_per_split` or we reach EOF.
+        let mut bytes_consumed = 0usize;
+        const BUFFER_SIZE: usize = 1024;
+        let mut buffer = [0u8; BUFFER_SIZE];
+        while bytes_consumed < self.bytes_per_split {
+            // Don't overshoot `self.bytes_per_split`!
+            let bytes_desired = std::cmp::min(BUFFER_SIZE, self.bytes_per_split - bytes_consumed);
+            let bytes_read = reader
+                .read(&mut buffer[0..bytes_desired])
+                .unwrap_or_else(|_| crash!(1, "error reading bytes from input file"));
+            // If we ever read 0 bytes then we know we've hit EOF.
+            if bytes_read == 0 {
+                return bytes_consumed;
+            }
+
+            writer
+                .write_all(&buffer[0..bytes_read])
+                .unwrap_or_else(|_| crash!(1, "error writing bytes to output file"));
+
+            bytes_consumed += bytes_read;
         }
-        self.bytes_to_write -= n;
-        if n == 0 {
-            self.bytes_to_write = self.saved_bytes_to_write;
-            control.request_new_file = true;
-        }
-        if self.break_on_line_end && n == line.chars().count() {
-            self.require_whole_line = self.break_on_line_end;
-        }
-        line[..n].to_owned()
+
+        bytes_consumed
     }
 }
 
@@ -363,14 +386,13 @@ fn split(settings: &Settings) -> i32 {
     let mut reader = BufReader::new(if settings.input == "-" {
         Box::new(stdin()) as Box<dyn Read>
     } else {
-        let r = match File::open(Path::new(&settings.input)) {
-            Ok(a) => a,
-            Err(_) => crash!(
+        let r = File::open(Path::new(&settings.input)).unwrap_or_else(|_| {
+            crash!(
                 1,
                 "cannot open '{}' for reading: No such file or directory",
                 settings.input
-            ),
-        };
+            )
+        });
         Box::new(r) as Box<dyn Read>
     });
 
@@ -380,48 +402,39 @@ fn split(settings: &Settings) -> i32 {
         a => crash!(1, "strategy {} not supported", a),
     };
 
-    let mut control = SplitControl {
-        current_line: "".to_owned(), // Request new line
-        request_new_file: true,      // Request new file
-    };
-
-    let mut writer = BufWriter::new(Box::new(stdout()) as Box<dyn Write>);
     let mut fileno = 0;
     loop {
-        if control.current_line.chars().count() == 0 {
-            match reader.read_line(&mut control.current_line) {
-                Ok(0) | Err(_) => break,
-                _ => {}
+        // Get a new part file set up, and construct `writer` for it.
+        let mut filename = settings.prefix.clone();
+        filename.push_str(
+            if settings.numeric_suffix {
+                num_prefix(fileno, settings.suffix_length)
+            } else {
+                str_prefix(fileno, settings.suffix_length)
             }
-        }
-        if control.request_new_file {
-            let mut filename = settings.prefix.clone();
-            filename.push_str(
-                if settings.numeric_suffix {
-                    num_prefix(fileno, settings.suffix_length)
-                } else {
-                    str_prefix(fileno, settings.suffix_length)
-                }
-                .as_ref(),
-            );
-            filename.push_str(settings.additional_suffix.as_ref());
+            .as_ref(),
+        );
+        filename.push_str(settings.additional_suffix.as_ref());
+        let mut writer = platform::instantiate_current_writer(&settings.filter, filename.as_str());
 
-            crash_if_err!(1, writer.flush());
-            fileno += 1;
-            writer = platform::instantiate_current_writer(&settings.filter, filename.as_str());
-            control.request_new_file = false;
-            if settings.verbose {
-                println!("creating file '{}'", filename);
+        let bytes_consumed = splitter.consume(&mut reader, &mut writer);
+        writer
+            .flush()
+            .unwrap_or_else(|e| crash!(1, "error flushing to output file: {}", e));
+
+        // If we didn't write anything we should clean up the empty file, and
+        // break from the loop.
+        if bytes_consumed == 0 {
+            // The output file is only ever created if filter's aren't used.
+            // Complicated, I know...
+            if settings.filter.is_none() {
+                remove_file(filename)
+                    .unwrap_or_else(|e| crash!(1, "error removing empty file: {}", e));
             }
+            break;
         }
 
-        let consumed = splitter.consume(&mut control);
-        crash_if_err!(1, writer.write_all(consumed.as_bytes()));
-
-        let advance = consumed.chars().count();
-        let clone = control.current_line.clone();
-        let sl = clone;
-        control.current_line = sl[advance..sl.chars().count()].to_owned();
+        fileno += 1;
     }
     0
 }
diff --git a/tests/by-util/test_split.rs b/tests/by-util/test_split.rs
index 521cbbe9a..37856f419 100644
--- a/tests/by-util/test_split.rs
+++ b/tests/by-util/test_split.rs
@@ -4,11 +4,15 @@ extern crate regex;
 use self::rand::{thread_rng, Rng};
 use self::regex::Regex;
 use crate::common::util::*;
+use rand::SeedableRng;
 #[cfg(not(windows))]
 use std::env;
-use std::fs::{read_dir, File};
 use std::io::Write;
 use std::path::Path;
+use std::{
+    fs::{read_dir, File},
+    io::BufWriter,
+};
 
 fn random_chars(n: usize) -> String {
     thread_rng()
@@ -58,7 +62,7 @@ impl Glob {
         files.sort();
         let mut data: Vec<u8> = vec![];
         for name in &files {
-            data.extend(self.directory.read(name).into_bytes());
+            data.extend(self.directory.read_bytes(name));
         }
         data
     }
@@ -81,20 +85,30 @@ impl RandomFile {
     }
 
     fn add_bytes(&mut self, bytes: usize) {
-        let chunk_size: usize = if bytes >= 1024 { 1024 } else { bytes };
-        let mut n = bytes;
-        while n > chunk_size {
-            let _ = write!(self.inner, "{}", random_chars(chunk_size));
-            n -= chunk_size;
+        // Note that just writing random characters isn't enough to cover all
+        // cases. We need truly random bytes.
+        let mut writer = BufWriter::new(&self.inner);
+
+        // Seed the rng so as to avoid spurious test failures.
+        let mut rng = rand::rngs::StdRng::seed_from_u64(123);
+        let mut buffer = [0; 1024];
+        let mut remaining_size = bytes;
+
+        while remaining_size > 0 {
+            let to_write = std::cmp::min(remaining_size, buffer.len());
+            let buf = &mut buffer[..to_write];
+            rng.fill(buf);
+            writer.write(buf).unwrap();
+
+            remaining_size -= to_write;
         }
-        let _ = write!(self.inner, "{}", random_chars(n));
     }
 
     /// Add n lines each of size `RandomFile::LINESIZE`
     fn add_lines(&mut self, lines: usize) {
         let mut n = lines;
         while n > 0 {
-            let _ = writeln!(self.inner, "{}", random_chars(RandomFile::LINESIZE));
+            writeln!(self.inner, "{}", random_chars(RandomFile::LINESIZE)).unwrap();
             n -= 1;
         }
     }
@@ -104,18 +118,18 @@ impl RandomFile {
 fn test_split_default() {
     let (at, mut ucmd) = at_and_ucmd!();
     let name = "split_default";
-    let glob = Glob::new(&at, ".", r"x[[:alpha:]][[:alpha:]]$");
     RandomFile::new(&at, name).add_lines(2000);
     ucmd.args(&[name]).succeeds();
+
+    let glob = Glob::new(&at, ".", r"x[[:alpha:]][[:alpha:]]$");
     assert_eq!(glob.count(), 2);
-    assert_eq!(glob.collate(), at.read(name).into_bytes());
+    assert_eq!(glob.collate(), at.read_bytes(name));
 }
 
 #[test]
 fn test_split_numeric_prefixed_chunks_by_bytes() {
     let (at, mut ucmd) = at_and_ucmd!();
     let name = "split_num_prefixed_chunks_by_bytes";
-    let glob = Glob::new(&at, ".", r"a\d\d$");
     RandomFile::new(&at, name).add_bytes(10000);
     ucmd.args(&[
         "-d", // --numeric-suffixes
@@ -123,52 +137,86 @@ fn test_split_numeric_prefixed_chunks_by_bytes() {
         "1000", name, "a",
     ])
     .succeeds();
+
+    let glob = Glob::new(&at, ".", r"a\d\d$");
     assert_eq!(glob.count(), 10);
-    assert_eq!(glob.collate(), at.read(name).into_bytes());
+    for filename in glob.collect() {
+        assert_eq!(glob.directory.metadata(&filename).len(), 1000);
+    }
+    assert_eq!(glob.collate(), at.read_bytes(name));
 }
 
 #[test]
 fn test_split_str_prefixed_chunks_by_bytes() {
     let (at, mut ucmd) = at_and_ucmd!();
     let name = "split_str_prefixed_chunks_by_bytes";
-    let glob = Glob::new(&at, ".", r"b[[:alpha:]][[:alpha:]]$");
     RandomFile::new(&at, name).add_bytes(10000);
+    // Important that this is less than 1024 since that's our internal buffer
+    // size. Good to test that we don't overshoot.
     ucmd.args(&["-b", "1000", name, "b"]).succeeds();
+
+    let glob = Glob::new(&at, ".", r"b[[:alpha:]][[:alpha:]]$");
     assert_eq!(glob.count(), 10);
-    assert_eq!(glob.collate(), at.read(name).into_bytes());
+    for filename in glob.collect() {
+        assert_eq!(glob.directory.metadata(&filename).len(), 1000);
+    }
+    assert_eq!(glob.collate(), at.read_bytes(name));
+}
+
+// This is designed to test what happens when the desired part size is not a
+// multiple of the buffer size and we hopefully don't overshoot the desired part
+// size.
+#[test]
+fn test_split_bytes_prime_part_size() {
+    let (at, mut ucmd) = at_and_ucmd!();
+    let name = "test_split_bytes_prime_part_size";
+    RandomFile::new(&at, name).add_bytes(10000);
+    // 1753 is prime and greater than the buffer size, 1024.
+    ucmd.args(&["-b", "1753", name, "b"]).succeeds();
+
+    let glob = Glob::new(&at, ".", r"b[[:alpha:]][[:alpha:]]$");
+    assert_eq!(glob.count(), 6);
+    for i in 0..5 {
+        assert_eq!(glob.directory.metadata(&glob.collect()[i]).len(), 1753);
+    }
+    assert_eq!(glob.directory.metadata(&glob.collect()[5]).len(), 1235);
+    assert_eq!(glob.collate(), at.read_bytes(name));
 }
 
 #[test]
 fn test_split_num_prefixed_chunks_by_lines() {
     let (at, mut ucmd) = at_and_ucmd!();
     let name = "split_num_prefixed_chunks_by_lines";
-    let glob = Glob::new(&at, ".", r"c\d\d$");
     RandomFile::new(&at, name).add_lines(10000);
     ucmd.args(&["-d", "-l", "1000", name, "c"]).succeeds();
+
+    let glob = Glob::new(&at, ".", r"c\d\d$");
     assert_eq!(glob.count(), 10);
-    assert_eq!(glob.collate(), at.read(name).into_bytes());
+    assert_eq!(glob.collate(), at.read_bytes(name));
 }
 
 #[test]
 fn test_split_str_prefixed_chunks_by_lines() {
     let (at, mut ucmd) = at_and_ucmd!();
     let name = "split_str_prefixed_chunks_by_lines";
-    let glob = Glob::new(&at, ".", r"d[[:alpha:]][[:alpha:]]$");
     RandomFile::new(&at, name).add_lines(10000);
     ucmd.args(&["-l", "1000", name, "d"]).succeeds();
+
+    let glob = Glob::new(&at, ".", r"d[[:alpha:]][[:alpha:]]$");
     assert_eq!(glob.count(), 10);
-    assert_eq!(glob.collate(), at.read(name).into_bytes());
+    assert_eq!(glob.collate(), at.read_bytes(name));
 }
 
 #[test]
 fn test_split_additional_suffix() {
     let (at, mut ucmd) = at_and_ucmd!();
     let name = "split_additional_suffix";
-    let glob = Glob::new(&at, ".", r"x[[:alpha:]][[:alpha:]].txt$");
     RandomFile::new(&at, name).add_lines(2000);
     ucmd.args(&["--additional-suffix", ".txt", name]).succeeds();
+
+    let glob = Glob::new(&at, ".", r"x[[:alpha:]][[:alpha:]].txt$");
     assert_eq!(glob.count(), 2);
-    assert_eq!(glob.collate(), at.read(name).into_bytes());
+    assert_eq!(glob.collate(), at.read_bytes(name));
 }
 
 // note: the test_filter* tests below are unix-only
@@ -182,15 +230,16 @@ fn test_filter() {
     // like `test_split_default()` but run a command before writing
     let (at, mut ucmd) = at_and_ucmd!();
     let name = "filtered";
-    let glob = Glob::new(&at, ".", r"x[[:alpha:]][[:alpha:]]$");
     let n_lines = 3;
     RandomFile::new(&at, name).add_lines(n_lines);
 
     // change all characters to 'i'
     ucmd.args(&["--filter=sed s/./i/g > $FILE", name])
         .succeeds();
+
     // assert all characters are 'i' / no character is not 'i'
     // (assert that command succeded)
+    let glob = Glob::new(&at, ".", r"x[[:alpha:]][[:alpha:]]$");
     assert!(
         glob.collate().iter().find(|&&c| {
             // is not i
@@ -209,7 +258,6 @@ fn test_filter_with_env_var_set() {
     // implemented like `test_split_default()` but run a command before writing
     let (at, mut ucmd) = at_and_ucmd!();
     let name = "filtered";
-    let glob = Glob::new(&at, ".", r"x[[:alpha:]][[:alpha:]]$");
     let n_lines = 3;
     RandomFile::new(&at, name).add_lines(n_lines);
 
@@ -217,7 +265,9 @@ fn test_filter_with_env_var_set() {
     env::set_var("FILE", &env_var_value);
     ucmd.args(&[format!("--filter={}", "cat > $FILE").as_str(), name])
         .succeeds();
-    assert_eq!(glob.collate(), at.read(name).into_bytes());
+
+    let glob = Glob::new(&at, ".", r"x[[:alpha:]][[:alpha:]]$");
+    assert_eq!(glob.collate(), at.read_bytes(name));
     assert!(env::var("FILE").unwrap_or("var was unset".to_owned()) == env_var_value);
 }