It is shown that the models for the transduction process in photoreceptors which treat latency and amplification as integrated phenomena (“integrated models”) yield time scales for single photon signals (“quantum bumps”) which distinctly conflict with the experimentally observed ones for the ventral nerve photoreceptor of Limulus: the ratio of bump duration/ latency t B/t lat is predicted by integrated models to be ≈3 in contrast to the experimental result of ≈0.5. Moreover, integrated models lead to a predicted value of an extinction rate of ≈50%, i.e., 50% of the absorbed photons should be expected to cause no signal in the dark adapted state of the cell. In this paper it is shown that separation of latency and amplification in such a way that the latency causing process precedes amplification in the transduction process eliminates these discrepancies. In addition, the separate modeling of latency and amplification resolves the rather large ambiguity in determining the exponent n of the initial signal current J(t)≈ nreported in the literature to be between n≈2 (from noise analysis) up to n≈17 (from flash experiments). Two alternative models for the latency part of transduction are suggested which give a qualitatively much better agreement with the experimental histograms of latencies.