The problems that arise when estimating the unknown parameters of models of spatial interaction are considered. Two models are analysed and the large-lattice theory outlined. The need for a small-lattice theory is discussed, and those aspects of the theory that distinguish it from the large-lattice theory are emphasized. The border-value problem, a central feature of small-lattice estimation theory, is discussed in relation to estimator bias. Properties of the least-squares and maximum-likelihood estimators in small-lattice situations are analysed.