Network structure has often proven to be important in understanding the decision behavior of individuals or agents for different interdependent situations. Computational studies predict that network structure has a crucial influence on behavior in iterated 2 by 2 asymmetric `battle of the sexes' games. We test such behavioral predictions in an experiment with 240 human subjects. We found that as expected the less `random' the network structure, the better the experimental results are predictable by the of the computational models. In particular, there is an effect of network clustering on the heterogeneity of convergence behavior in the network. We also found that degree centrality and having an even degree are important predictors for the decision behavior of the subjects in the experiment. We thus find empirical validation of previous computational models using human subjects.