Equivalence-equivalence responding (Eq-Eq) has become a behaviour analytic model of analogical reasoning. In previous works it was demonstrated that the exposition to a non-arbitrary relational task (facilitation procedure) improves performance in Eq-Eq tasks. In the present work we attempted to analyze the role of task components: arbitrary or non-arbitrary relational responses, role as sample or comparisons, and relating relations. In the first experiment, we devised four facilitation procedures combining two dimensions: simple or compound sample or comparisons and arbitrary or non-arbitrary relations among compound stimuli. In the second experiment two facilitation procedures including compound stimuli were tested. In one condition arbitrary relations worked as sample, and non-arbitrary relations as comparison. In the other condition its function was reversed. All procedures were effective to improve Eq-Eq to different extents, being arbitrary relational responses the key element. These results show generalization between non-arbitrary and arbitrary responses, and add further support to Eq-Eq responding as operant behaviour.