diff --git a/gcs/gcs.go b/gcs/gcs.go
index b5c4fb8..ccf73e3 100644
--- a/gcs/gcs.go
+++ b/gcs/gcs.go
@@ -6,6 +6,7 @@
 package gcs
 
 import (
+	"encoding/binary"
 	"fmt"
 	"io"
 	"sort"
@@ -140,6 +141,24 @@ func FromBytes(N uint32, P uint8, d []byte) (*Filter, error) {
 	return f, nil
 }
 
+// FromNBytes deserializes a GCS filter from a known P, and serialized N and
+// filter as returned by NBytes().
+func FromNBytes(P uint8, d []byte) (*Filter, error) {
+	return FromBytes(binary.BigEndian.Uint32(d[:4]), P, d[4:])
+}
+
+// FromPBytes deserializes a GCS filter from a known N, and serialized P and
+// filter as returned by NBytes().
+func FromPBytes(N uint32, d []byte) (*Filter, error) {
+	return FromBytes(N, d[0], d[1:])
+}
+
+// FromNPBytes deserializes a GCS filter from a serialized N, P, and filter as
+// returned by NPBytes().
+func FromNPBytes(d []byte) (*Filter, error) {
+	return FromBytes(binary.BigEndian.Uint32(d[:4]), d[4], d[5:])
+}
+
 // Bytes returns the serialized format of the GCS filter, which does not
 // include N or P (returned by separate methods) or the key used by SipHash.
 func (f *Filter) Bytes() []byte {
@@ -148,6 +167,34 @@ func (f *Filter) Bytes() []byte {
 	return filterData
 }
 
+// NBytes returns the serialized format of the GCS filter with N, which does
+// not include P (returned by a separate method) or the key used by SipHash.
+func (f *Filter) NBytes() []byte {
+	filterData := make([]byte, len(f.filterData)+4)
+	binary.BigEndian.PutUint32(filterData[:4], f.n)
+	copy(filterData[4:], f.filterData)
+	return filterData
+}
+
+// PBytes returns the serialized format of the GCS filter with P, which does
+// not include N (returned by a separate method) or the key used by SipHash.
+func (f *Filter) PBytes() []byte {
+	filterData := make([]byte, len(f.filterData)+1)
+	filterData[0] = f.p
+	copy(filterData[1:], f.filterData)
+	return filterData
+}
+
+// NPBytes returns the serialized format of the GCS filter with N and P, which
+// does not include the key used by SipHash.
+func (f *Filter) NPBytes() []byte {
+	filterData := make([]byte, len(f.filterData)+5)
+	binary.BigEndian.PutUint32(filterData[:4], f.n)
+	filterData[4] = f.p
+	copy(filterData[5:], f.filterData)
+	return filterData
+}
+
 // P returns the filter's collision probability as a negative power of 2 (that
 // is, a collision probability of `1/2**20` is represented as 20).
 func (f *Filter) P() uint8 {
diff --git a/gcs/gcs_test.go b/gcs/gcs_test.go
index 56d8568..de31bd8 100644
--- a/gcs/gcs_test.go
+++ b/gcs/gcs_test.go
@@ -6,6 +6,7 @@
 package gcs_test
 
 import (
+	"bytes"
 	"encoding/binary"
 	"math/rand"
 	"testing"
@@ -23,7 +24,7 @@ var (
 	// Filters are conserved between tests but we must define with an
 	// interface which functions we're testing because the gcsFilter
 	// type isn't exported
-	filter, filter2 *gcs.Filter
+	filter, filter2, filter3, filter4, filter5 *gcs.Filter
 
 	// We need to use the same key for building and querying the filters
 	key [gcs.KeySize]byte
@@ -90,6 +91,18 @@ func TestGCSFilterCopy(t *testing.T) {
 	if err != nil {
 		t.Fatalf("Filter copy failed: %s", err.Error())
 	}
+	filter3, err = gcs.FromNBytes(filter.P(), filter.NBytes())
+	if err != nil {
+		t.Fatalf("Filter copy failed: %s", err.Error())
+	}
+	filter4, err = gcs.FromPBytes(filter.N(), filter.PBytes())
+	if err != nil {
+		t.Fatalf("Filter copy failed: %s", err.Error())
+	}
+	filter5, err = gcs.FromNPBytes(filter.NPBytes())
+	if err != nil {
+		t.Fatalf("Filter copy failed: %s", err.Error())
+	}
 }
 
 // TestGCSFilterMetadata checks that the filter metadata is built and
@@ -104,9 +117,39 @@ func TestGCSFilterMetadata(t *testing.T) {
 	if filter.P() != filter2.P() {
 		t.Fatal("P doesn't match between copied filters")
 	}
+	if filter.P() != filter3.P() {
+		t.Fatal("P doesn't match between copied filters")
+	}
+	if filter.P() != filter4.P() {
+		t.Fatal("P doesn't match between copied filters")
+	}
+	if filter.P() != filter5.P() {
+		t.Fatal("P doesn't match between copied filters")
+	}
 	if filter.N() != filter2.N() {
 		t.Fatal("N doesn't match between copied filters")
 	}
+	if filter.N() != filter3.N() {
+		t.Fatal("N doesn't match between copied filters")
+	}
+	if filter.N() != filter4.N() {
+		t.Fatal("N doesn't match between copied filters")
+	}
+	if filter.N() != filter5.N() {
+		t.Fatal("N doesn't match between copied filters")
+	}
+	if !bytes.Equal(filter.Bytes(), filter2.Bytes()) {
+		t.Fatal("Bytes don't match between copied filters")
+	}
+	if !bytes.Equal(filter.Bytes(), filter3.Bytes()) {
+		t.Fatal("Bytes don't match between copied filters")
+	}
+	if !bytes.Equal(filter.Bytes(), filter4.Bytes()) {
+		t.Fatal("Bytes don't match between copied filters")
+	}
+	if !bytes.Equal(filter.Bytes(), filter5.Bytes()) {
+		t.Fatal("Bytes don't match between copied filters")
+	}
 }
 
 // TestGCSFilterMatch checks that both the built and copied filters match