In this letter, we investigate distributed detection over channels with memory. We model the wireless channel between each sensor and the fusion center by a two state Markov chain and derive the optimal fusion rule. We demonstrate the improved detection performance of the derived fusion rule for burst error channels, over a channel aware fusion rule for a memoryless channel and the Chair-Varshney fusion rule. We further show that when the number of sensor decisions is large, as well as when the channel is memoryless, the derived fusion rule reduces to the well known counting rule.