Development of a time-temperature-based food quality prediction using self-organized network is described. The self-organized network is combined with a back-propagation output layer to form a three layer prediction neural netwrok. A method for automatic construction of neurons in hidden layer inspired by immune algorithm is proposed. The prediction system is tested using real meat delivery data form Akita to Chiba, Japan. The experiment shows that the system recognizes the trained data from 93.3% up to 100% and outperforms the recognition obtained by conventional back-propagation network which is from 73.3% up to 80%.