Exact predictions for two-pulse visual temporal integration data are derived from the Bouman-van der Velden quantum coincidence model for threshold vision. The predictions of the model start with complete summation for superposed pulses, then pass to a transition zone of partial integration, and finally reach the level of probability summation for pulses presented with large interstimulus intervals. From our results we can clearly reject the assumption of constant integration times with the basic model. We thus generalize the coincidence model to allow for variable integration times, derive the corresponding predictions for two-pulse integration data, and compare these predictions to published data currently available. It is shown that detectors of low order of coincidence generally underestimate the actual reduction of threshold intensity (or equivalently the corresponding increase of the detection probability) for two pulses as compared to the simple-pulse performance.