diff --git a/subsys/net/ip/net_pkt.c b/subsys/net/ip/net_pkt.c
index 501ee43d5963f..94b0a9a6c90be 100644
--- a/subsys/net/ip/net_pkt.c
+++ b/subsys/net/ip/net_pkt.c
@@ -1163,6 +1163,8 @@ u16_t net_pkt_append(struct net_pkt *pkt, u16_t len, const u8_t *data,
 		    s32_t timeout)
 {
 	struct net_buf *frag;
+	struct net_context *ctx;
+	size_t max_len;
 
 	if (!pkt || !data) {
 		return 0;
@@ -1177,6 +1179,40 @@ u16_t net_pkt_append(struct net_pkt *pkt, u16_t len, const u8_t *data,
 		net_pkt_frag_add(pkt, frag);
 	}
 
+	/* Make sure we don't send more data in one packet than
+	 * MTU allows.
+	 * This check really belongs to net_pkt_append_bytes(),
+	 * but instead done here for efficiency, we assume
+	 * (at least for now) that that callers of
+	 * net_pkt_append_bytes() are smart enough to not
+	 * overflow MTU.
+	 */
+	max_len = 0x10000;
+
+	ctx = net_pkt_context(pkt);
+	if (ctx) {
+#if defined(CONFIG_NET_TCP)
+		if (ctx->tcp) {
+			max_len = ctx->tcp->send_mss;
+		} else
+#endif
+		{
+			struct net_if *iface = net_context_get_iface(ctx);
+			max_len = net_if_get_mtu(iface);
+			/* Optimize for number of jumps in the code ("if"
+			 * instead of "if/else").
+			 */
+			max_len -= NET_IPV4TCPH_LEN;
+			if (net_context_get_family(ctx) != AF_INET) {
+				max_len -= NET_IPV6TCPH_LEN - NET_IPV4TCPH_LEN;
+			}
+		}
+	}
+
+	if (len > max_len) {
+		len = max_len;
+	}
+
 	return net_pkt_append_bytes(pkt, data, len, timeout);
 }
 
diff --git a/subsys/net/lib/sockets/sockets.c b/subsys/net/lib/sockets/sockets.c
index a2f59e15fdd42..dd7d0b845dc94 100644
--- a/subsys/net/lib/sockets/sockets.c
+++ b/subsys/net/lib/sockets/sockets.c
@@ -235,7 +235,6 @@ ssize_t zsock_sendto(int sock, const void *buf, size_t len, int flags,
 	struct net_pkt *send_pkt;
 	s32_t timeout = K_FOREVER;
 	struct net_context *ctx = INT_TO_POINTER(sock);
-	size_t max_len = net_if_get_mtu(net_context_get_iface(ctx));
 
 	ARG_UNUSED(flags);
 
@@ -249,18 +248,6 @@ ssize_t zsock_sendto(int sock, const void *buf, size_t len, int flags,
 		return -1;
 	}
 
-	/* Make sure we don't send more data in one packet than
-	 * MTU allows. Optimize for number of branches in the code.
-	 */
-	max_len -= NET_IPV4TCPH_LEN;
-	if (net_context_get_family(ctx) != AF_INET) {
-		max_len -= NET_IPV6TCPH_LEN - NET_IPV4TCPH_LEN;
-	}
-
-	if (len > max_len) {
-		len = max_len;
-	}
-
 	len = net_pkt_append(send_pkt, len, buf, timeout);
 	if (!len) {
 		net_pkt_unref(send_pkt);