In this paper we recast the problem of missing values in the covariates of a regression model as a latent Gaussian Markov random field (GMRF)