The present work relates to detection of the freshness of fish by using electronic nose composed of 8 metal oxide sensors. The advantages of the decision tree structure are applied to evaluate fish freshness by selecting both classification algorithm and extracted features in every node of the decision tree structure. Additionally, storage conditions of fish are different from other works about fish freshness. In this work, storage conditions of fisherman are taken into consideration. Thus, to determine fish freshness is become difficult. Also the sub-sampling method used as feature extraction method is increased the classification algorithm success. The proposed algorithm is composed of combination of support vector machine (SVM) and k-Nearest Neighbour (k-NN) method in decision tree structure. It is shown that the proposed algorithm is able to distinguish fish freshness to seven classes with success rate as 97.22%. Also this algorithm is compared with k-NN and artificial neural network.