A crucial issue in all-optical networks is packet loss. In this paper we evaluate sources of packet loss, comparing impact of effects at the physical layer and at the network layer. The study is compiled for optical packet, burst and circuit switched networks. We provide an analytical model that evaluates packet loss due to bit errors using bit-error rate, packet length distribution and network size as parameters. Bit errors at the physical layer set a lower limit to the aggregate packet loss and for some scenarios it overshadows packet loss at the network layer. For applications applied in the Internet today, bit error requirements may be considerably alleviated as compared to those of synchronous digital hierarchy (SDH) systems without degrading the perceived quality of service for the end-user. By considering recommended packet loss rates for future Internet services we evaluate the usefulness of different techniques for reduced packet loss.