One of the consequences of the low specificity of the in vitro mammalian cell genotoxicity assays reported in our previous paper [D. Kirkland, M. Aardema, L. Henderson, L. Muller, Evaluation of the ability of a battery of three in vitro genotoxicity tests to discriminate rodent carcinogens and non-carcinogens. I. Sensitivity, specificity and relative predictivity, Mutat. Res. 584 (2005) 1-256] is industry and regulatory agencies dealing with a large number of false-positive results during the safety assessment of new chemicals and drugs. Addressing positive results from in vitro genotoxicity assays to determine which are "false" requires extensive resources, including the conduct of additional animal studies. In order to reduce animal usage, and to conserve industry and regulatory agency resources, we thought it was important to raise the question as to whether the protocol requirements for a valid in vitro assay or the criteria for a positive result could be changed in order to increase specificity without a significant loss in sensitivity of these tests. We therefore analysed some results of the mouse lymphoma assay (MLA) and the chromosomal aberration (CA) test obtained for rodent carcinogens and non-carcinogens in more detail. For a number of chemicals that are positive only in either of these mammalian cell tests (i.e. negative in the Ames test) there was no correlation between rodent carcinogenicity and level of toxicity (we could not analyse this for the CA test as insufficient data were available in publications), magnitude of response or lowest effective positive concentration. On the basis of very limited in vitro and in vivo data, we could also find no correlation between the above parameters and formation of DNA adducts. Therefore, a change to the current criteria for required level of toxicity in the MLA, to limit positive calls to certain magnitudes of response, or to certain concentration ranges would not improve the specificity of the tests without significantly reducing the sensitivity. We also investigated a possible correlation between tumour profile (trans-species, trans-sex and multi-site versus single-species, single-sex and single-site) and pattern of genotoxicity results. Carcinogens showing the combination of trans-species, trans-sex and multi-site tumour profile were much more prevalent (70% more) in the group of chemicals giving positive results in all three in vitro assays than amongst those giving all negative results. However, single-species, single-sex, single-site carcinogens were not very prevalent even amongst those chemicals giving three negative results in vitro. Surprisingly, when mixed positive and negative results were compared, multi-site carcinogens were highly prevalent amongst chemicals giving only a single positive result in the battery of three in vitro tests. Finally we extended our relative predictivity (RP) calculations to combinations of positive and negative results in the genotoxicity battery. For two out of three tests positive, the RP for carcinogenicity was no higher than 1.0 and for 2/3 tests negative the RP for non-carcinogenicity was either zero (for Ames+MLA+MN) or 1.7 (for Ames+MLA+CA). Thus, all values were less than a meaningful RP of two, and indicate that it is not possible to predict outcome of the rodent carcinogenicity study when only 2/3 genotoxicity results are in agreement.