VAN HEERDEN has discussed some of the differences between his holographic model of memory1 and a pair of non-holographic models that we put forward last year2. We alluded to the poor signal to noise ratio of the holograph when it is used to reconstruct a stored pattern from a fragment of that pattern; van Heerden's comment refers to the signal to noise ratio with which one can locate a given fragment in a large text, and this is quite a different matter. Van Heerden himself showed that the signal to noise ratio in the reconstruction of random patterns was equal to the size of the fragment divided by the size of the whole pattern, and the same applies to the reconstruction of a temporal signal from a short cue3.