Government regulations and the search for PREPS have resulted in an unprecedented amount of data and data analyses. Data reduction using ANOVA followed by a posteriori tests is useful for single variables, but a multivariate approach is preferable for correlated variables such as smoke yields of tar, nicotine, CO, formaldehyde, HCN and benzene (Canadian 'Pack Labels') and mutagenic responses in the Ames test (Health Canada test method T-501). To illustrate the approach, 'Pack Label' data was obtained for 17 brands of Canadian cigarettes smoked under ISO conditions. Per cigarette yields ranged from 10-16 mg tar and CO, 0.9-1.2 mg nicotine, 45-105 µg formaldehyde, 85-160 µg HCN and 40-60 µg benzene. With the exception of nicotine, all correlation coefficients for 'Pack Label' analytes were significant with r = 0.5. Univariate analysis resulted in 6 ANOVA tables while a single table summarized the results from the multivariate approach. In both cases, it was concluded that yields of tar, CO and HCN, taken together, had the best discriminating power. By way of illustration, the multivariate technique was also applied to typical results from the Ames test. Here, the lack of correlation among the responses of the 5 stains resulted in a significant loss of information in the multivariate approach. Consequently, for this type of data, univariate analysis is preferable since it does not result in the loss of information. However, for correlated variables, multivariate analysis is recommended since it reduces complexity and takes into account possible interactions.