Gaussian Markov random fields (GMRFs) are an important example of MRFs with many applications, particularly because GMRFs are known to provide effective approximations to any MRF. Despite their relative computational simplicity, inference in GMRFs using maximum likelihood (ML) is not always feasible. Therefore, this paper compares the inference quality using the pseudolikelihood, a well-known computational shortcut to full ML, and in addition the generalized lambda distribution is simulated to investigate robustness to departure from the Gaussian distribution.