Recent research has shown how high spatial resolution auxiliary data may improve sampling for field estimates. This paper defines theoretically the variance and bias of field estimates resulting from sampling with a model-based and a mean-based inference. The factors influencing bias and variance are identified and their respective impacts quantified. The theoretical formalisation demonstrates how the characteristics of the auxiliary data sampled and their representativeness affect the mean estimation. Both model-based and mean-based approaches are then tested and compared with real and simulated data. The results highlight the differences between model-based and mean-based for estimating mean yield field values. This paper highlights the value of using a model-based approach when auxiliary data are available. In most cases, results are notably better than the use of common mean-based inference.