diff --git a/src/encoding/csv/fuzz.go b/src/encoding/csv/fuzz.go
deleted file mode 100644
index 5f5cdfcbf8..0000000000
--- a/src/encoding/csv/fuzz.go
+++ /dev/null
@@ -1,70 +0,0 @@
-// Copyright 2019 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-//go:build gofuzz
-
-package csv
-
-import (
-	"bytes"
-	"fmt"
-	"reflect"
-)
-
-func Fuzz(data []byte) int {
-	score := 0
-	buf := new(bytes.Buffer)
-
-	for _, tt := range []Reader{
-		{},
-		{Comma: ';'},
-		{Comma: '\t'},
-		{LazyQuotes: true},
-		{TrimLeadingSpace: true},
-		{Comment: '#'},
-		{Comment: ';'},
-	} {
-		r := NewReader(bytes.NewReader(data))
-		r.Comma = tt.Comma
-		r.Comment = tt.Comment
-		r.LazyQuotes = tt.LazyQuotes
-		r.TrimLeadingSpace = tt.TrimLeadingSpace
-
-		records, err := r.ReadAll()
-		if err != nil {
-			continue
-		}
-		score = 1
-
-		buf.Reset()
-		w := NewWriter(buf)
-		w.Comma = tt.Comma
-		err = w.WriteAll(records)
-		if err != nil {
-			fmt.Printf("writer  = %#v\n", w)
-			fmt.Printf("records = %v\n", records)
-			panic(err)
-		}
-
-		r = NewReader(buf)
-		r.Comma = tt.Comma
-		r.Comment = tt.Comment
-		r.LazyQuotes = tt.LazyQuotes
-		r.TrimLeadingSpace = tt.TrimLeadingSpace
-		result, err := r.ReadAll()
-		if err != nil {
-			fmt.Printf("reader  = %#v\n", r)
-			fmt.Printf("records = %v\n", records)
-			panic(err)
-		}
-
-		if !reflect.DeepEqual(records, result) {
-			fmt.Println("records = \n", records)
-			fmt.Println("result  = \n", records)
-			panic("not equal")
-		}
-	}
-
-	return score
-}
diff --git a/src/encoding/csv/fuzz_test.go b/src/encoding/csv/fuzz_test.go
new file mode 100644
index 0000000000..6342fa416d
--- /dev/null
+++ b/src/encoding/csv/fuzz_test.go
@@ -0,0 +1,96 @@
+// Copyright 2019 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package csv
+
+import (
+	"bytes"
+	"reflect"
+	"slices"
+	"strings"
+	"testing"
+)
+
+func FuzzRoundtrip(f *testing.F) {
+	f.Fuzz(func(t *testing.T, in []byte) {
+		buf := new(bytes.Buffer)
+
+		t.Logf("input = %q", in)
+		for _, tt := range []Reader{
+			{Comma: ','},
+			{Comma: ';'},
+			{Comma: '\t'},
+			{Comma: ',', LazyQuotes: true},
+			{Comma: ',', TrimLeadingSpace: true},
+			{Comma: ',', Comment: '#'},
+			{Comma: ',', Comment: ';'},
+		} {
+			t.Logf("With options:")
+			t.Logf("  Comma            = %q", tt.Comma)
+			t.Logf("  LazyQuotes       = %t", tt.LazyQuotes)
+			t.Logf("  TrimLeadingSpace = %t", tt.TrimLeadingSpace)
+			t.Logf("  Comment          = %q", tt.Comment)
+			r := NewReader(bytes.NewReader(in))
+			r.Comma = tt.Comma
+			r.Comment = tt.Comment
+			r.LazyQuotes = tt.LazyQuotes
+			r.TrimLeadingSpace = tt.TrimLeadingSpace
+
+			records, err := r.ReadAll()
+			if err != nil {
+				continue
+			}
+			t.Logf("first records = %#v", records)
+
+			buf.Reset()
+			w := NewWriter(buf)
+			w.Comma = tt.Comma
+			err = w.WriteAll(records)
+			if err != nil {
+				t.Logf("writer  = %#v\n", w)
+				t.Logf("records = %v\n", records)
+				t.Fatal(err)
+			}
+			if tt.Comment != 0 {
+				// Writer doesn't support comments, so it can turn the quoted record "#"
+				// into a non-quoted comment line, failing the roundtrip check below.
+				continue
+			}
+			t.Logf("second input = %q", buf.Bytes())
+
+			r = NewReader(buf)
+			r.Comma = tt.Comma
+			r.Comment = tt.Comment
+			r.LazyQuotes = tt.LazyQuotes
+			r.TrimLeadingSpace = tt.TrimLeadingSpace
+			result, err := r.ReadAll()
+			if err != nil {
+				t.Logf("reader  = %#v\n", r)
+				t.Logf("records = %v\n", records)
+				t.Fatal(err)
+			}
+
+			// The reader turns \r\n into \n.
+			for _, record := range records {
+				for i, s := range record {
+					record[i] = strings.ReplaceAll(s, "\r\n", "\n")
+				}
+			}
+			// Note that the reader parses the quoted record "" as an empty string,
+			// and the writer turns that into an empty line, which the reader skips over.
+			// Filter those out to avoid false positives.
+			records = slices.DeleteFunc(records, func(record []string) bool {
+				return len(record) == 1 && record[0] == ""
+			})
+			// The reader uses nil when returning no records at all.
+			if len(records) == 0 {
+				records = nil
+			}
+
+			if !reflect.DeepEqual(records, result) {
+				t.Fatalf("first read got %#v, second got %#v", records, result)
+			}
+		}
+	})
+}