diff --git a/index.js b/index.js
index 545536f..dd65c3c 100644
--- a/index.js
+++ b/index.js
@@ -20,7 +20,7 @@ const defaults = {
 
 const DANGEROUS_KEYS = new Set(['__proto__', 'constructor', 'prototype'])
 
-function sanitizeHeader(header) {
+function sanitizeHeader (header) {
   if (typeof header !== 'string') {
     return null
   }
@@ -56,6 +56,7 @@ class CsvParser extends Transform {
       escaped: false,
       first: true,
       lineNumber: 0,
+      pendingCr: false,
       previousEnd: 0,
       rowLength: 0,
       quoted: false
@@ -241,6 +242,12 @@ class CsvParser extends Transform {
 
     if (this._prev) {
       start = this._prev.length
+      if (this.state.pendingCr) {
+        // Revisit the CR now that its following byte may be available.
+        start--
+        this.state.rowLength--
+        this.state.pendingCr = false
+      }
       buffer = Buffer.concat([this._prev, data])
       this._prev = null
     }
@@ -274,6 +281,10 @@ class CsvParser extends Transform {
           if (chr === nl) {
             this.options.newline = nl
           } else if (chr === cr) {
+            if (nextChr === null) {
+              this.state.pendingCr = true
+              break
+            }
             if (nextChr !== nl) {
               this.options.newline = cr
             }
diff --git a/test/newline-chunks.test.js b/test/newline-chunks.test.js
new file mode 100644
index 0000000..55a6ee2
--- /dev/null
+++ b/test/newline-chunks.test.js
@@ -0,0 +1,92 @@
+const test = require('ava')
+const csv = require('..')
+
+function parse (chunks, options) {
+  return new Promise((resolve, reject) => {
+    const parser = csv(options)
+    const rows = []
+    let headers
+    parser.on('headers', (value) => { headers = value })
+    parser.on('data', (row) => rows.push(row))
+    parser.on('error', reject)
+    parser.on('end', () => resolve({ headers, rows }))
+    for (const chunk of chunks) parser.write(Buffer.from(chunk))
+    parser.end()
+  })
+}
+
+test('CRLF detection waits for the next chunk (#234)', async (t) => {
+  const expected = { headers: ['foo'], rows: [{ foo: 'bar' }, { foo: 'baz' }] }
+  t.deepEqual(await parse(['foo\r\nbar\r\nbaz\r\n']), expected)
+  t.deepEqual(await parse(['foo\r', '\nbar\r\n', 'baz\r\n']), expected)
+})
+
+test('CRLF records and byte offsets do not depend on chunk boundaries', async (t) => {
+  const input = 'name,value\r\nCleo,1\r\nPancakes,2\r\n'
+  const options = { outputByteOffset: true, strict: true }
+  const expected = {
+    headers: ['name', 'value'],
+    rows: [
+      { row: { name: 'Cleo', value: '1' }, byteOffset: 12 },
+      { row: { name: 'Pancakes', value: '2' }, byteOffset: 20 }
+    ]
+  }
+  for (const split of [4, 11, 12, 19]) {
+    t.deepEqual(await parse([input.slice(0, split), input.slice(split)], options), expected)
+  }
+  t.deepEqual(await parse(input.split(''), options), expected)
+})
+
+test('empty chunks do not resolve a pending CR', async (t) => {
+  t.deepEqual(await parse(['foo\r', '', '', '\nbar']), {
+    headers: ['foo'], rows: [{ foo: 'bar' }]
+  })
+})
+
+test('standalone CR is detected when the next byte arrives', async (t) => {
+  t.deepEqual(await parse(['foo\r', 'bar\r', 'baz\r']), {
+    headers: ['foo'], rows: [{ foo: 'bar' }, { foo: 'baz' }]
+  })
+})
+
+test('EOF handles a pending CR and an unterminated final row', async (t) => {
+  t.deepEqual(await parse(['foo\r']), { headers: ['foo'], rows: [] })
+  t.deepEqual(await parse(['foo']), { headers: ['foo'], rows: [] })
+  for (const ending of ['', '\r', '\r\n']) {
+    t.deepEqual(await parse(['foo\r', `\nbar${ending}`]), {
+      headers: ['foo'], rows: [{ foo: 'bar' }]
+    })
+  }
+})
+
+test('LF and explicit custom newline behavior is preserved', async (t) => {
+  for (const newline of ['\n', '\r', 'X']) {
+    t.deepEqual(await parse([`foo${newline}`, `bar${newline}`], { newline }), {
+      headers: ['foo'], rows: [{ foo: 'bar' }]
+    })
+  }
+})
+
+test('quoted CRLF stays in the header and field', async (t) => {
+  t.deepEqual(await parse(['"fo\r', '\no"\r', '\n"bar\r', '\nbaz"\r', '\n']), {
+    headers: ['fo\r\no'], rows: [{ 'fo\r\no': 'bar\r\nbaz' }]
+  })
+})
+
+test('skipped lines and comments can end with a split CRLF', async (t) => {
+  t.deepEqual(await parse(['skip\r', '\nfoo\r', '\nbar\r\n'], { skipLines: 1 }), {
+    headers: ['foo'], rows: [{ foo: 'bar' }]
+  })
+  t.deepEqual(await parse(['#comment\r', '\nfoo\r', '\nbar\r\n'], { skipComments: true }), {
+    headers: ['foo'], rows: [{ foo: 'bar' }]
+  })
+})
+
+test('a deferred CR counts once towards maxRowBytes', async (t) => {
+  const chunks = ['foo\r', '\nbar\r\n']
+  t.deepEqual(await parse(chunks, { maxRowBytes: 5 }), {
+    headers: ['foo'], rows: [{ foo: 'bar' }]
+  })
+  await t.throwsAsync(parse(chunks, { maxRowBytes: 4 }), { message: 'Row exceeds the maximum size' })
+  await t.throwsAsync(parse(['foo\r'], { maxRowBytes: 3 }), { message: 'Row exceeds the maximum size' })
+})