Accurate Forest type maps over large areas are important to construct an optimal decision support tool for forest ecosystems management. Recently, digital classification of remotely sensed data has been increasingly used as a suitable replacement to produce these maps. The obtained results are different based on the classification methods, forest covers, study area conditions and satellite data. This research utilizes data from SPOT5 images to compare the results of pixel-based, object-oriented and decision tree classification methods for mapping forest types in the Astara forests located in the northern parts of Iran. To obtain the field information, we chose a random systematic sampling a approach, where 153 plots of 1 hectare were sampled. Forest type in each plot was identified based on the percent of frequency of different species including the mixed Parrotia, mixed hardwood, mixed Carpinus, mixed Fagus and protected lands. In order to map these forest types, we used a pixel-based neural network classifier method. Forest types could be classified using spectral data with an overall accuracy of 52.04% and a kappa coefficient of 0.39. In addition to the pixel-based approach, the object-oriented classification method consisting of segmentation of image data into the primary objects and employment of the Nearest neighbor and membership functions was used for classication. The overall accuracy and kappa coefficient of classification reached 63.3% and 0.54 respectivelly by using the spectral data alone, Results of the accuracy assessment showed that, the performance of the object-oriented method for discrimination of forest types is considerably better than the pixel-based method. Spectral signatures of forest types showed overlaps and most of them were porrly separable in the feature space provided by remotely sensed data. Therefore, to improve discrimination of forest types. We used a texture measure and ancillary data such as DEM, slope. Aspect and distance to stream data which could be effective for mapping spatial occurrence of mixed forest types. Integration of the ancillary data with spectral data in a decision tree classifier resulted in the improvement of the classification results. In this process, overall accuracy of 76.5% and a kappa coefficient of 0.7 were obtained and DEM was the most useful layer for forest type separation and Comparison in the results of pixel-based and object-oriented classification methods showed that the objectoriented method has higher capabilities for classifying forest types. Also, higher accuracies could be achieved for separation of mixed forest types by the decision tree classifier.