The purpose of this investigation was to establish relationships between the colour of cigarette filters, measured after smoking, and the yield of mandated analytes (Health Canada). Three cigarette brands (2 flue-cured and 1 blended) were smoked using 6 smoking regimens selected to cover a range of human smoking behaviours (smoke volume per cigarette; 225 to 700 ml). Yields of 40 analytes were determined for each condition as well as filter efficiency for nicotine. In a separate series of experiments, filter colour (after smoking) was measured spectrophotometrically (CIELab) both at the mouth end and the rod end under two simulated light conditions (daylight and fluorescent). Typical estimates of precision (relative standard deviations) for the colour measurements were L* (black/white), 2.9%; a* (red/green), 8.7%; and b*(yellow/blue), 5.1%. Multiple regression models relating L*, a* and b* to analyte yield were constructed for a subset of 20 compounds based on the colour of rod end of the filter using daylight illumination. Stepwise regression was used to select the best model. Ten of these involved a* alone; 7 involved L* alone; and 3 involved both a* and b*. Also, there were 10 examples where the addition of a term for filter efficiency resulted in a significant increase in the explained variation (R2). Average R2 for the complete models, taken over all 20 regressions, was 0.85 ranging from 0.60 for NO to 0.97 for nicotine. When the models were used to estimate amounts delivered by cigarettes, coefficients of variation (CV) ranged from approximately 8% (1,3-butadiene) to slightly over 50% (NNK). The CV for tar was 14.8%. A simple non invasive filter colour measurement may be helpful in estimating exposure to tar and other tobacco smoke constituents.