Probability distributions and different mixed regression models are used to study of wet and dry spells. In this study, eight stations on the southern coast of the Caspian Sea were used for 55 years (1960 to 2015). One and two day wet spells are the most frequent in the area. Also, four and higher days wet spells in the western region is more than the east. The best probability distributions of wet spells are two-parameter gamma, two parameter normal log, two parameter logistic log, four-parameter generalized gamma, generalized three parameters gamma, two-parameter gamma, three-parameter gamma and a three-parameter logistic log. Also, the dry spells of the region are better fitted with generalized Parto, generalized logistics and generalized extreme value distributions. Due to the random behavior of the annual wet and dry spells, the best generalized linear mixed model for wet spells is the negative binomial log, Log Poisson, log normal and gamma logarithmic, also the Log Poisson, log normal and the negative binomial distribution was determined for the annual dry spells.