Existing studies on spatial panel data models typically assume a normal distribution for the random error components. This assumption may not be appropriate in many applications. Here we consider a more flexible and powerful approach that generalizes the traditional model. We propose a skew-normal generalized spatial panel data model that adopts a multivariate skew normal distribution for the random error components. For parameter estimation, a Bayesian inference algorithm is developed. A simulation study and the analysis of a real data set of cigarette demand are conducted to compare the proposed skew normal spatial model with the traditional (normal) spatial model.