Conventional stochastic frontier models often impose the assumptions that the errors are cross-sectionally independent. However, they are unrealistic in many applications with panel data. This paper considers a linear model with time-invariant fixed effects to represent heterogeneity, cross-sectional dependence by introducing common correlated effects, and the time-variant technical inefficiency and idiosyncratic errors jointly characterized by a multivariate skew normal distribution. To consistently estimate the slope coefficients and variances in the above model, we propose a transformation to eliminate fixed effects and common correlated effects. Based on the transformed likelihood function, we then introduce an EM Algorithm to robustly estimate these parameters. Our Monte Carlo simulation shows that the proposed method is quite accurate in the presence of common correlated effects, while conventional models without taking these effects into account can result in severally biased parameter estimates.