diff --git a/bencode/api.go b/bencode/api.go
index b5f026c5..1b77f588 100644
--- a/bencode/api.go
+++ b/bencode/api.go
@@ -129,7 +129,7 @@ func Unmarshal(data []byte, v interface{}) error {
 }
 
 func NewDecoder(r io.Reader) *Decoder {
-	return &Decoder{r: bufio.NewReader(r)}
+	return &Decoder{r: &scanner{r: r}}
 }
 
 func NewEncoder(w io.Writer) *Encoder {
diff --git a/bencode/decode_test.go b/bencode/decode_test.go
index 1db0c7e4..572740b8 100644
--- a/bencode/decode_test.go
+++ b/bencode/decode_test.go
@@ -72,16 +72,28 @@ func TestDecoderConsecutive(t *testing.T) {
 
 func TestDecoderConsecutiveDicts(t *testing.T) {
 	bb := bytes.NewBufferString("d4:herp4:derped3:wat1:ke17:oh baby a triple!")
+
 	d := NewDecoder(bb)
+	assert.EqualValues(t, "d4:herp4:derped3:wat1:ke17:oh baby a triple!", bb.Bytes())
+	assert.EqualValues(t, 0, d.offset)
+
 	var m map[string]interface{}
+
 	require.NoError(t, d.Decode(&m))
 	assert.Len(t, m, 1)
 	assert.Equal(t, "derp", m["herp"])
+	assert.Equal(t, "d3:wat1:ke17:oh baby a triple!", bb.String())
+	assert.EqualValues(t, 14, d.offset)
+
 	require.NoError(t, d.Decode(&m))
 	assert.Equal(t, "k", m["wat"])
+	assert.Equal(t, "17:oh baby a triple!", bb.String())
+	assert.EqualValues(t, 24, d.offset)
+
 	var s string
 	require.NoError(t, d.Decode(&s))
 	assert.Equal(t, "oh baby a triple!", s)
+	assert.EqualValues(t, 44, d.offset)
 }
 
 func check_error(t *testing.T, err error) {
diff --git a/bencode/scanner.go b/bencode/scanner.go
new file mode 100644
index 00000000..eaf22dd4
--- /dev/null
+++ b/bencode/scanner.go
@@ -0,0 +1,41 @@
+package bencode
+
+import (
+	"errors"
+	"io"
+)
+
+// Implements io.ByteScanner over io.Reader, for use in Decoder, to ensure
+// that as little as the undecoded input Reader is consumed as possible.
+type scanner struct {
+	r      io.Reader
+	b      [1]byte // Buffer for ReadByte
+	unread bool    // True if b has been unread, and so should be returned next
+}
+
+func (me *scanner) Read(b []byte) (int, error) {
+	return me.r.Read(b)
+}
+
+func (me *scanner) ReadByte() (byte, error) {
+	if me.unread {
+		me.unread = false
+		return me.b[0], nil
+	}
+	n, err := me.r.Read(me.b[:])
+	if err != nil {
+		return me.b[0], err
+	}
+	if n != 1 {
+		panic(n)
+	}
+	return me.b[0], err
+}
+
+func (me *scanner) UnreadByte() error {
+	if me.unread {
+		return errors.New("byte already unread")
+	}
+	me.unread = true
+	return nil
+}