This work presents a novel fish classification methodology based on a robust feature selection technique. Unlike existing works for fish classification, which propose feature descriptors and do not analyze their individual impacts in the whole classification task. A problem with classification of fish species is still vital facets due to: arbitrary fish size and orientation; feature variability; environmental changes; poor image quality; segmentation failures; imaging conditions; physical shaping; distortion; noise; overlap, and occlusion of objects in digital images. In addition, the problem in fish classification is to find meaningful features based on the image segmentation and features extraction, and an efficient classifier that produces a better fish images classification accuracy rate. Thus, this research aims to design and develop a novel fish classifier based on an appropriate feature set obtained from image segmentation and features extraction methods, to classify the given fish output into its cluster (poison and non-poison fish), therefore; classifying the clustered poison and non-poison fish into its family.