Incompletely documented symptom episodes pose methodological problems in the analysis of diary data. The aim of this study was to develop a method of estimating the average durations of symptomatic and nonsymptomatic episodes, respectively, coping with the problem of bias due to undocumented days and censored episodes that is found in most diary studies. The authors derived their outcome variables from a Markov model using transition probabilities. To evaluate this method, the authors assessed the impact of active smoking on the duration of episodes of bronchitis symptoms and the corresponding nonsymptomatic periods, respectively, using diary data (1992-1993) obtained from 801 participants in the Swiss Study on Air Pollution and Lung Diseases in Adults. Covariate-adjusted distribution curves for the mean durations of individual episodes were estimated by Cox regression. Median values for light smokers (<10 cigarettes/day) were 60.0 symptom-free days (95% confidence interval (CI) 42.0-78.5) and 4.0 symptomatic days (95% CI 3.0-6.0), respectively, compared with medians of only 21.0 days (95% CI 16.2-29.8) for periods without bronchitis symptoms and 6.0 days (95% CI 4.9-9.0) for episodes of bronchitis symptoms in heavy smokers (> or =30 cigarettes/day). The authors suggest that the Markov method is a feasible approach to the assessment of long term effects of smoking and environmental risk factors on the average duration of symptomatic and nonsymptomatic respiratory episodes.