Always Mazandaran province had been exposed to development and changes because of location in coastline of the Caspian Sea and having unique conditions. GIS and remote sensing can be used to monitor in order to manage this sensitive province. Then, this study is trying to detect land cover changes in coastal areas of mazandaran province using LCM. Land cover changes detection have done using Landsat satellite images belonging to the years 1988, 2001, 2006, and 2010. Modeling the transition potential was performed by using multi-layer perceptron artificial neural network and 8 variables. Then modeling was done for 2010 by using the hard predict model and 2001-2006 calibration period and in order to determine accuracy of evaluate it was compared with ground truth map in 2010 year. Finally, land cover in 2016 was predicted by using the 2006-2010 calibration periods. The results showed that the total period of study 33487 ha of forest area has declined. And 21367 and 13155 ha has been added to the extent of agricultural and residential lands, respectively. More changes forest related to conversion to agriculture (30424 ha) and then converted to residential (1265 ha). In most of sub-models, the results of potential modeling of using artificial neural networks demonstrated high accuracy (52-94 percent). The total error in the modeling was obtained 12.84% for 2010, which this reflects high compliance in predicted image by model with the ground truth image. Modeling results for 2016 showed that, area of forest and open land will be lower compared to the 2010 (9988 and 429ha, respectively) and agricultural and residential land increased (respectively 7607 and 2810 hectares).