It is possible to trace the phenological profile of targets on the Earth’s surface through multitemporal remote sensing data. Different features can be computed from multitemporal data to classify land use classes. In this context, this paper presents a new method to map the land use based on the probabilistic analysis of multitemporal features using Bayesian Networks. Elementary statistical measures were computed from NDVI/MODIS and EVI/MODIS time series of pasture, sugarcane, annual agriculture and other uses classes for 2012/2013 and 2013/2014 crop years in southern Goiás state, Brazil. The model’s output is composed by layers representing the occurrence probability of each class over the study area. A thematic map was built from output layers and the classification was evaluated by the Monte Carlo simulation. In our preliminary results, we obtained classification accuracy values within Kappa index range from 0.51 to 0.63. Annual agriculture and other land use classes were more easily distinguished and more confusion happened between pasture and sugarcane classes. Although the accuracy values were not high, the proposed model presented a potential for land use classification and it can be improved.