Abstarct: An artificial neural network (ANN) model in quantitative structure property relationship (QSPR) was developed for density prediction of 6 ketones over a wide range of pressure molecular (2-300MPa) and temperatures (273.15-338.15 K). A large number of descxriptors were calculated by Dragon software and a subset of calculated descxriptors was selected from 18 classes of Dragon descxriptors with a stepwise multiple linear regression (MLR) as a feature selection technique. The selected descxriptors that appear in multiple linear regression models are: 2D-ATS4e (weigthed by atomic sanderson electronegativities) and RDF040e (weigthed by atomic sanderson electronegativites). Two calculated and two experimental descxriptors contain: pressure and temperature, were selected as the most feasible descxriptors in the construction of artificial neural network (ANN) models. The data set was randomly divided into three subsets: training (324 point), validation (107 point) and test set (107 point). After training and optimization of the ANN parameters, the performance of the model was investigated by the test set. The mean squares error (MSE) were 0.9360 and 0.1735 respectively, for the test data set in MLR and ANN methods. The results obtained using ANN were compared with the experimental values as well as with those obtained using regression models and showed the superiority of ANN over linear multiple regression model.