This paper proposes a stochastic frontier panel data model in which unit-specific inefficiencies are spatially correlated. In particular, this model has simultaneously three important features: i) the total inefficiency of a productive unit depends on its own inefficiency and on the inefficiency of its neighbors; ii) the spatially correlated and time varying inefficiency is disentangled from time invariant unobserved heterogeneity in a panel data model a la Greene (2005); iii) systematic differences in inefficiency can be explained using exogenous determinants. We propose to estimate both the “true” fixed- and random-effects variants of the model using a feasible simulated composite maximum likelihood approach. The finite sample behavior of the proposed estimators are investigated through a set of Monte Carlo experiments. Our simulation results suggest that the estimation approach is consistent, showing good finite sample properties especially in small samples.