One of the characteristics of precipitation is its long-term changes over time. Considering the complexities of precipitation behavior, using nonlinear methods can lead to a better understanding of this phenomenon. Therefore, in this study, an attempt has been made to investigate the annual precipitation trend in northwestern Iran in the statistical period 1970 to 2013 using polynomial models along with the linear method. Discrete wavelet transform has also been used to reduce the effect of variability on the trend. The time series used to survey the trend is obtained from the grid data of the daily precipitation of northwestern Iran from 1/1/1970 to 12/31/2013 (16071 days). These data are the result of interpolation of daily precipitation using kriging method in lambert conformal conic coordinate system. The result showed that the long-term behavior of precipitation in northwestern Iran in the period under study follows a 3 degree pattern; in a way that the mean frequency of annual precipitation in the first decade (1970) had been increasing rapidly. Since the first years of the 80's, a significant downward trend in the data had been formed and continued for nearly 3 decades. However, in the final years, an upward trend had begun again. The results also showed that data decomposition with wavelet transform while maintaining the mean and long-term behavior in the approximate component has also filtered precipitation variability from this component. Hence the low frequency component obtained from series decomposition is a good representative for trend estimation.