Microbial records of commercial raw milk (total counts) and two ice creams (standard plate counts, coliforms and yeasts) were de-rounded. Their autocorrelation function showed that the counts were independent and had no discernible trend or periodicity. The first half of each record was used to estimate the frequencies of high counts in the second half, assuming that the counts have a lognormal, log-Laplace or non-parametric distribution. The estimated frequencies were all on the same order of magnitude of those actually observed and in some cases in close agreement. None of the three distributions was found to be consistently superior to the others, most probably because the available records for the analysis were too short. In general, the lognormal and log-Laplace distributions yielded closer estimates than the non-parametric distribution because they allow for high counts at levels not previously recorded. The results strengthen the view that at least, in principle, the irregular fluctuating pattern of the microbial records of dairy products can be translated into a set of probabilities of future high counts, and that these can serve as a risk or quality measure.