Two different approaches to the design of optimal observations networks are compared. One approach is based on the traditional experimental design theory, the other essentially uses the covariance analysis methodology of observed fields, It is found that for random fields generated by regression models with random parameters both approaches lead to similar solutions