A method for determining a matrix of expression levels corresponding to a set of biological targets (e.g., genes or gene fragments) and a set of biological samples, including obtaining a matrix of signal values corresponding to the set of biological targets; computing a vector of expression levels for a sample in the set of biological samples using the matrix of signal values; storing the vector of computed expression levels in a storage matrix; repeating the computing and storing steps for each sample in the set of biological samples; and outputting the storage matrix as the matrix of expression levels. The method, based on a linear programming formulation of the problem, works for both "promiscuous" probe array data, in which there may be multiple targets indicated by a single probe, and the "polygamous" case, in which there are multiple probes for a single target. The preferred method can also process data obtained from multiple SAGE analyses using multiple markers. A second embodiment of the method determines optimal expression levels when the available probe data is noisy or uncertain.