diff --git a/bench_test.go b/bench_test.go
new file mode 100644
index 0000000..c8e28d2
--- /dev/null
+++ b/bench_test.go
@@ -0,0 +1,58 @@
+package taskgroup_test
+
+import (
+	"math/rand"
+	"sync"
+	"testing"
+)
+
+// A very rough benchmark comparing the performance of accumulating values with
+// a separate goroutine via a channel, vs. accumulating them directly under a
+// lock. The workload here is intentionally minimal, so the benchmark is
+// measuring more or less just the overhead.
+
+func BenchmarkChan(b *testing.B) {
+	ch := make(chan int)
+	done := make(chan struct{})
+	var total int
+	go func() {
+		defer close(done)
+		for v := range ch {
+			total += v
+		}
+	}()
+	b.ResetTimer() // discount the setup time.
+
+	var wg sync.WaitGroup
+	wg.Add(b.N)
+	for i := 0; i < b.N; i++ {
+		go func() {
+			defer wg.Done()
+			ch <- rand.Intn(1000)
+		}()
+	}
+	wg.Wait()
+	close(ch)
+	<-done
+}
+
+func BenchmarkLock(b *testing.B) {
+	var μ sync.Mutex
+	var total int
+	report := func(v int) {
+		μ.Lock()
+		defer μ.Unlock()
+		total += v
+	}
+	b.ResetTimer() // discount the setup time.
+
+	var wg sync.WaitGroup
+	wg.Add(b.N)
+	for i := 0; i < b.N; i++ {
+		go func() {
+			defer wg.Done()
+			report(rand.Intn(1000))
+		}()
+	}
+	wg.Wait()
+}