A two-state Markov chain model provided very serviceable approximations to the April-September frequency statistics for duration of dry spells recorded in 45 years' observations at Edmonton, Swift Current and Winnipeg. It was less satisfactory in respect of the total number of rainy days per month, tending to underestimate the frequency of months with few rainy days. There was a minor degree of correlation in the numbres of rainy days recorded during the same month at these three locations, but no significant statistical association in those for successive months at the same location. Abstract A two-state Markov chain model provided very serviceable approximations to the April-September frequency statistics for duration of dry spells recorded in 45 years' observations at Edmonton, Swift Current and Winnipeg. It was less satisfactory in respect of the total number of rainy days per month, tending to underestimate the frequency of months with few rainy days. There was a minor degree of correlation in the numbres of rainy days recorded during the same month at these three locations, but no significant statistical association in those for successive months at the same location.