An automated method for counting spot-forming units in the ELISpot assay is described that uses a statistical model fit to training data that is based on counts from one or more experts. The method adapts to variable background intensities and provides considerable flexibility with respect to what image features can be used to model expert counts. Point estimates of spot counts are produced together with intervals that reflect the degree of uncertainty in the count. Finally, the approach is completely transparent and "open source" in contrast to methods embedded in current commercial software. An illustrative application to data from a study of the reactivity of T-cells from healthy human subjects to a pool of immunodominant peptides from CMV, EBV and flu is presented.