Petrophysical evaluation is a key stage before formation test in exploration field, it allows the identification of the best plays in term of effective porosity, shale volume and hydrocarbon saturation. This paper introduces an innovative approach to petrophysical evaluation by leveraging deep learning techniques, our scheme is based on three streams, in each stream we predict a petrophysical parameter (Volume of shale, Effective Porosity and water saturation), each stream represents a convolutional neural network model that has been trained using traditional wireline logging data from the Silurien argilo-greseux reservoir units in the Berkine Basin, Algeria. experimental results show that our proposed method achieves stat-of-the-art predictions by giving correlated results in comparison against conventional analysis methods (calculated Shale Volume, Porosity and water saturation), furthermore assessment of test data shows also that our method gives good prediction in thin beds reservoirs and offers the ability to detect low resistivity pays. Because our method is based on Convolutional neural network models, it is fast and allows the petrophysical parameters prediction in real time.