A Horvitz-Thompson predictor is proposed for spatial sampling when the characteristic of interest is modeled as a random field. Optimal sampling designs are deduced under this context. Fixed and variable sample size are considered.