The proton affinities of some primary, secondary, and tertiary amines have been calculated with different semiempirical and nonempirical quantum chemical methods. We were particularly interested in the question which of the most popular semiempirical methods yield good overall correlations between calculated and experimental values and, therefore, allow a reliable prediction of hitherto unknown proton affinities. We found that some of the most frequently used semiempirical methods result in good correlations only within the groups of primary, secondary and tertiary amines, while the overall correlation is even worse than the one obtained with the noniterative EHT method. Among the more recent methods which allow geometry optimizations (MINDO/3, MNDO, AMI, PM3, MSINDO) the best results have been calculated with the MSINDO method. Testing for the influence of geometry optimization we surprisingly found that two of these methods (MINDO/3, AMI) perform even better when geometry optimizations are omitted and standard bonding parameters are used instead. Superior results, however, have been obtained with the CNDO/2- and the INDO method. Finally, the best correlations between semiempirically calculated and experimental proton affinities have been achieved with the spectroscopic parametrizations of these methods, CNDO/2S and INDO/2S, respectively. The correlations resulting in these cases are close to those reached at the ZPE+MP2/6-31 l++G**//HF/6-311++G** level of