Abstarct: A quantitative structure-property relationship (QSPR) approach was employed to predict the density of associative fluids such as alcohols, primary and secondary amines and carboxylic acids. Two different methods were used to generat the significant descxriptors. In the first method, a large number of descxriptors was calculated using Dragon software and the best calculated descxriptors was selected from 18 classes of Dragon descxriptors by stepwise regression (SR) and genetic algorithm baxsed on partial least square (GA-PLS) methods. 8 descxriptors with SR and 5 descxriptors with GA-PLS methods were selected. In the second method, the best descxriptors are selected baxsed on group contribution method. After analyzing the chemical structures of all compounds in this work, 10 functionally groups (-CH3, -CH2- , >CH-, >CH-OH, , -COOH, -CH2-NH2, >CH-NH2, >NH) were found useful. The selected descxriptors with three feature selection methods and two experimental descxriptors (pressure and temperature) were used as input nodes for constructing different artificial networks. 4 feed forward artificial neural networks were designed by combination of Levenberg-Marquardt and Bayesian regularized (BR) algorithms with logarithm-sigmoid and tangent-sigmoid transfer functions. After training and optimization of the ANNs parameters such as, number of neurons in input and hidden laxyers, and number of iteration (epoch), the performance of the models was investigated by the test set. The results showed that the Levenberg-Marquart artificial neural network with tangent-sigmoid transfer function and GCM baxsed descxriptors, may be simulated the relationship between the structural descxriptors and density of the desired molecules accurately. The mean square errors (MSEs) and Absolute Avarage precent Deviation (AADs) for the test set are 1.2432 and 0.0867%, respectively.