Abstract: In order to make the artificial immune algorithms more efficient, learning mechanisms which obtain experiences from population's previous evolution and use them as a guide for further developments are very important for this kind of stochastic searching algorithms. To address this, Memetic algorithms combine local search heuristics with evolutionary operators and their learning mechanisms make a decision about which heuristic to be suitable for the target problem. However, users have to provide several problem dependent heuristics in advance. In this paper, an unsupervised learning mechanism based on ant colony pheromone is designed for the function optimization problem. Based on the proposed learning mechanism, a novel Memetic algorithm termed pheromone meme based clonal selection algorithm (PM_CSA) is also put forward. In MP_CSA, the pheromone is used as a carrier of evolutionary experiences, and the concentration distribution of the pheromone acts as a guider for generating new individuals. Different from conventional Memetic algorithms, the pheromone based learning is not to make a choice of which predefined meme will be employed but to obtain developing experience from evolution. Experimental results indicate that the pheromone based learning has the ability of acquiring useful information about the objective functions. It significantly improves the performance of standard clonal selection algorithm.