To investigate the underlying mechanisms which produce self-generation effect from a network theory of memory, two experiments were undertaken. In Experiment 1, Ss produced 10 persons' names which had positive affect and they were given 10 names which had neutral affect by an experimenter. Then Ss constructed sentences using the names which were generated or given with an object (noun) and its modifier. In Experiment 2, the same procedure was adopted, however, this time Ss produced persons' names which had negative affect. Incidental free and cued recall tests showed strong self-generation effects in both experiments. These results were explained in terms of higher activation of the nodes in the memory network. Higher recall results of Experiment 1 in every recall condition compared to those of Experiment 2, were explained in terms of memory organization differences depending on the difference of affects between two Experiments. The self-generation effect in the modifier recall were explained by spreading activation from the related node, such as nouns. The organizational aspect of the memory was also explained in terms of the degree of the connection of nodes in the network.