Pinpointing salt distribution in seismic interpretation and hydrocarbon exploration plays a major role, so this study focused on supervised learning in salt detection. The seismic data utilized is the North Sea F3 block and in the location of the Zechstein evaporite group. Five attributes selected (polar dip, dip variance, most positive curvature, similarity, semblance) which were qualified for differentiating salt from non-salt data. Feature selection included sampling, two-class labeling (salt and non-salt), and identifying the importance of features by implementing bagged decision trees. Finally, tree attributes consist of dip variance, most positive curvature, and similarity selected and input to the learning process using a multi-layer perceptron neural network (MLP). The results showed the power of MLP in following patterns introduced by the interpreter, and consequently, the importance of feature selection in supervised learning.