In Denmark, a national serological surveillance-and-control programme for Salmonella in pigs has been in operation since 1995. The programme is based on the Danish mix-ELISA and uses double testing (two ELISA-wells used per sample) of meat-juice samples taken in relation to slaughter. All herds are classified monthly into one of the three levels; the classification is based on the percentage of positive serological results in the previous 3 months. In connection with evaluation of the programme in 2001, we investigated whether single testing (testing in one well only) could be expected to be sufficiently precise compared to double testing. Data from the year 2000 were used, and mathematical modelling. Single testing was simulated by randomised selection of one of the two results in the double testing. A slight increase in the prevalence of Salmonella-positive samples (1.02-1.09 times more through the four quarters of the year 2000) was found in the simulated single testing, as compared to the double testing. Around 0.5% of the herds would be allocated to another herd level in single testing-almost equal numbers one level up and one level down. No herd being seronegative in double testing would be allocated to levels 2 or 3 (herds with >40 or >70%, respectively, serological reactors) in single testing. The prevalence of "false-positive" diagnoses (positive in single testing and negative in double testing) and inversely defined "false-negative" diagnoses varied from 4.2 to 8.7% and from 3.2 to 4.5%, respectively, through the four quarters of the year 2000. The probability of allocating a herd to a wrong level due to sampling error was on the average 6.2 (varying from 1.66 to over 100) times higher than the probability of allocating a herd to a wrong level due to the test inaccuracy introduced by going from double to single testing. This is, however, an average; a herd with a true prevalence close to one of the level border cut-offs (40 and 70% weighted seroprevalence, respectively) would have a higher risk of being allocated to a wrong level than a herd with a true prevalence far from the level border cut-offs. The results are based on the current Danish sample sizes in the surveillance scheme, which implies that 60, 75 or 100 samples are taken annually in a herd, depending on its size. Other sample sizes would produce other results.