This paper considers discrete survey sampling in an area, when it can be assumed that observations near each other will be positively correlated. We consider two methods of analyzing such data: classic randomized complete block design (RCBD), and random field linear models (RFLMs) where the spatial covariance function is estimated directly from the data. When normality can be assumed, RFLM reduces to the likelihood analysis for the model Y∼N[Xβ,Σ(θ)]. The mean E(Y)=Xβ is the standard linear model formulation, and the covariance function Σ(θ) is modeled using the exponential, Gaussian, or spherical formulation from geostatistics. Hypothesis tests for β were formulated using standard RCBD-ANOVA; and likelihood ratio and Wald statistics were used for the RFLM model. Simulations indicate that for blocked experimental designs both RCBD-ANOVA and RFLM perform well. For experimental designs in which adequate blocking is unavailable, simulations indicate standard ANOVA can perform poorly, but RFLM may be a useful tool to improve the quality of inference that can be derived. The RCBD-ANOVA and RFLM were applied to catch data from a sea scallop survey experiment in the New York Bight conducted by the Northeast Fisheries Science Center using three research vessels. Both RCBD-ANOVA and RFLM analysis showed there was no significant difference in the fishing power of these vessels. [Copyright &y& Elsevier]