diff --git a/notify.go b/notify.go
deleted file mode 100644
index f560d398..00000000
--- a/notify.go
+++ /dev/null
@@ -1,75 +0,0 @@
-package pubsub
-
-import (
-	"github.com/libp2p/go-libp2p/core/network"
-	"github.com/libp2p/go-libp2p/core/peer"
-	ma "github.com/multiformats/go-multiaddr"
-)
-
-var _ network.Notifiee = (*PubSubNotif)(nil)
-
-type PubSubNotif PubSub
-
-func (p *PubSubNotif) OpenedStream(n network.Network, s network.Stream) {
-}
-
-func (p *PubSubNotif) ClosedStream(n network.Network, s network.Stream) {
-}
-
-func (p *PubSubNotif) Connected(n network.Network, c network.Conn) {
-	// ignore transient connections
-	if c.Stat().Limited {
-		return
-	}
-
-	go func() {
-		p.newPeersPrioLk.RLock()
-		p.newPeersMx.Lock()
-		p.newPeersPend[c.RemotePeer()] = struct{}{}
-		p.newPeersMx.Unlock()
-		p.newPeersPrioLk.RUnlock()
-
-		select {
-		case p.newPeers <- struct{}{}:
-		default:
-		}
-	}()
-}
-
-func (p *PubSubNotif) Disconnected(n network.Network, c network.Conn) {
-}
-
-func (p *PubSubNotif) Listen(n network.Network, _ ma.Multiaddr) {
-}
-
-func (p *PubSubNotif) ListenClose(n network.Network, _ ma.Multiaddr) {
-}
-
-func (p *PubSubNotif) Initialize() {
-	isTransient := func(pid peer.ID) bool {
-		for _, c := range p.host.Network().ConnsToPeer(pid) {
-			if !c.Stat().Limited {
-				return false
-			}
-		}
-
-		return true
-	}
-
-	p.newPeersPrioLk.RLock()
-	p.newPeersMx.Lock()
-	for _, pid := range p.host.Network().Peers() {
-		if isTransient(pid) {
-			continue
-		}
-
-		p.newPeersPend[pid] = struct{}{}
-	}
-	p.newPeersMx.Unlock()
-	p.newPeersPrioLk.RUnlock()
-
-	select {
-	case p.newPeers <- struct{}{}:
-	default:
-	}
-}
diff --git a/peer_notify.go b/peer_notify.go
new file mode 100644
index 00000000..44aceeef
--- /dev/null
+++ b/peer_notify.go
@@ -0,0 +1,112 @@
+package pubsub
+
+import (
+	"context"
+
+	"github.com/libp2p/go-libp2p/core/event"
+	"github.com/libp2p/go-libp2p/core/network"
+	"github.com/libp2p/go-libp2p/core/peer"
+	"github.com/libp2p/go-libp2p/core/protocol"
+)
+
+func (ps *PubSub) watchForNewPeers(ctx context.Context) {
+	// We don't bother subscribing to "connectivity" events because we always run identify after
+	// every new connection.
+	sub, err := ps.host.EventBus().Subscribe([]interface{}{
+		&event.EvtPeerIdentificationCompleted{},
+		&event.EvtPeerProtocolsUpdated{},
+	})
+	if err != nil {
+		log.Errorf("failed to subscribe to peer identification events: %v", err)
+		return
+	}
+	defer sub.Close()
+
+	ps.newPeersPrioLk.RLock()
+	ps.newPeersMx.Lock()
+	for _, pid := range ps.host.Network().Peers() {
+		if ps.host.Network().Connectedness(pid) != network.Connected {
+			continue
+		}
+		ps.newPeersPend[pid] = struct{}{}
+	}
+	ps.newPeersMx.Unlock()
+	ps.newPeersPrioLk.RUnlock()
+
+	select {
+	case ps.newPeers <- struct{}{}:
+	default:
+	}
+
+	var supportsProtocol func(protocol.ID) bool
+	if ps.protoMatchFunc != nil {
+		var supportedProtocols []func(protocol.ID) bool
+		for _, proto := range ps.rt.Protocols() {
+
+			supportedProtocols = append(supportedProtocols, ps.protoMatchFunc(proto))
+		}
+		supportsProtocol = func(proto protocol.ID) bool {
+			for _, fn := range supportedProtocols {
+				if (fn)(proto) {
+					return true
+				}
+			}
+			return false
+		}
+	} else {
+		supportedProtocols := make(map[protocol.ID]struct{})
+		for _, proto := range ps.rt.Protocols() {
+			supportedProtocols[proto] = struct{}{}
+		}
+		supportsProtocol = func(proto protocol.ID) bool {
+			_, ok := supportedProtocols[proto]
+			return ok
+		}
+	}
+
+	for ctx.Err() == nil {
+		var ev any
+		select {
+		case <-ctx.Done():
+			return
+		case ev = <-sub.Out():
+		}
+
+		var protos []protocol.ID
+		var peer peer.ID
+		switch ev := ev.(type) {
+		case event.EvtPeerIdentificationCompleted:
+			peer = ev.Peer
+			protos = ev.Protocols
+		case event.EvtPeerProtocolsUpdated:
+			peer = ev.Peer
+			protos = ev.Added
+		default:
+			continue
+		}
+
+		// We don't bother checking connectivity (connected and non-"limited") here because
+		// we'll check when actually handling the new peer.
+
+		for _, p := range protos {
+			if supportsProtocol(p) {
+				ps.notifyNewPeer(peer)
+				break
+			}
+		}
+	}
+
+}
+
+func (ps *PubSub) notifyNewPeer(peer peer.ID) {
+	ps.newPeersPrioLk.RLock()
+	ps.newPeersMx.Lock()
+	ps.newPeersPend[peer] = struct{}{}
+	ps.newPeersMx.Unlock()
+	ps.newPeersPrioLk.RUnlock()
+
+	select {
+	case ps.newPeers <- struct{}{}:
+	default:
+	}
+}
diff --git a/pubsub.go b/pubsub.go
index c4ecae65..24c297dd 100644
--- a/pubsub.go
+++ b/pubsub.go
@@ -327,14 +327,12 @@ func NewPubSub(ctx context.Context, h host.Host, rt PubSubRouter, opts ...Option
 			h.SetStreamHandler(id, ps.handleNewStream)
 		}
 	}
-	h.Network().Notify((*PubSubNotif)(ps))
+	go ps.watchForNewPeers(ctx)
 
 	ps.val.Start(ps)
 
 	go ps.processLoop(ctx)
 
-	(*PubSubNotif)(ps).Initialize()
-
 	return ps, nil
 }
 
@@ -687,6 +685,8 @@ func (p *PubSub) handlePendingPeers() {
 	p.newPeersPrioLk.Unlock()
 
 	for pid := range newPeers {
+		// Make sure we have a non-limited connection. We do this late because we may have
+		// disconnected in the meantime.
 		if p.host.Network().Connectedness(pid) != network.Connected {
 			continue
 		}