Collective memory is a mechanism through which a community of agents can learn improved patterns of cooperative activity to take advantage of regularities in their domain of problem-solving activity. We have previously shown how a collective memory of cooperative procedures can be used to reduce the effort, and, in this paper, we demonstrated how a CM of COBWEB trees can be used to predict operator probabilities as an impoverished model of agent capabilities. Finally, we discussed how this model of behavior suits our domain of interest better than related work.