In an infinitely repeated matching pennies game with asymmetric information, Gossner et al. (2006) studied coordination levels among agents and obtained that players may use a block codification using signaling mistakes in order to efficiently coordinate. Inspired in that model, we experimentally test coordination in the lab. We first model and establish the appropriate length of the sequence played by nature and the block strategy for a finitely repeated version of the game, where the majority rule with 3-length blocks results as the optimal block codification. Our experimental data give support to the main results of the original model with respect to the codification rule using signaling mistakes.