This research investigates the efficiency of number of neurons used in self-organizing mapping, one of the artificial neural network models, on detecting construct of scales. In this method, the construct of a scale can differ as the number of neurons used for mapping changes. In this study, the methods for determining the optimum number of neurons to detect the construct of a scale are confirmatory factor analysis and distinct groups method. The research findings reveal that increasing neuron numbers causes reduction through one factor of the scale construct and, at the same time, the fit and error indexes of this scale constructed in one factor have more fitted model than those of others. In addition to this, the evidence from distinct groups method supports this finding. As a result, it is recommended that the number of neurons should increase till the related items of a scale are gathered in a neuron in the method of self-organizing mapping used for construct validity. In addition to this rule, it is recommended to analyze the revealed construct with respect to related attitude variable contextually.