Results of serological monitoring for Salmonella in finishing pigs are used to classify herds and target control measures at herds with high prevalence. The outcome of monitoring is determined by three factors: (a) the cut-off value for the optical density percentage (OD%) to declare a sample positive, (b) the classification scheme to allocate farms to different Salmonella prevalence classes, and (c) the annual number of samples per herd to calculate its Salmonella prevalence. Our goal was to analyse the impact of these three factors on (i) the accuracy of Salmonella monitoring in finishing pigs and (ii) the total number of tests required. We constructed a stochastic simulation model in Excel and @Risk to evaluate 12 monitoring scenarios based on: (a) four cut-off values for the OD% (10, 20, 30, and 40) and (b) three herd classification schemes. Furthermore, eight different sampling schemes were evaluated. The main outputs of the model are (a) the accuracy of monitoring which is reflected by the percentage of herds that retain classification when resampled at the same moment in time and (b) the total number of tests. To illustrate the model, we used input data from Salmonella monitoring in Lower Saxony, Germany. Model calculations demonstrated that - with the tests in use - monitoring scenarios based on cut-off OD% 10 are most accurate with 80-90% of herds retaining classification. Monitoring scenarios based on cut-off OD% 20 or 30 are, however, comparable to those based on cut-off OD% 40 with 50-70% of herds retaining classification. Besides, we predicted that herd classifications based on three classes (low-, moderate-, and high-prevalence) give more accurate results than when a zero-prevalence class is included. The total number of tests depends heavily on the sampling scheme and - if sampling is based on Salmonella prevalence class - the distribution of herds over the different classes. We predicted that the current German sampling scheme that is based on herd size requires more tests than those sampling schemes based on herd classification. Of these, the sampling scheme in which most samples are taken from high-prevalence herds is most accurate and might be a good incentive to reduce Salmonella prevalence at herd level if farmers had to pay for the tests themselves. (C) 2007 Elsevier B.V. All rights reserved.