One of the main outstanding problems in keyword mnemotechnics is whether this technique is more effective when the subjects generate their own keywords, or when the keywords are supplied by the experimenter. Both methods have advantages and disadvantages. An alternative method has recently been suggested, in which the keywords are generated by the subjects' peers. In the present study we aimed to investigate whether immediate or delayed recall are affected by keyword generation method (experimenter or peer generation). We also aimed to determine whether the method of keyword generation affects imaging capacity as evaluated by questionnaires or spatial tests. A total of 377 secondary-school students were selected and divided into four groups. All subjects were presented with 30 Latin words. Additionally, the subjects in Group 1 were presented with bizarre images, while the subjects of Group 2 were presented with normal images, in both cases representing the keywords selected by peers as interacting with the Latin words. The subjects in Groups 3 and 4 were likewise presented with normal or bizarre images, respectively, but representing the keywords selected by the experimenters. The subjects' imaging capacity was evaluated by means of the Vividness of Visual Imagery Questionnaire (VVIQ) and the Spatial Test of Primary Mental Abilities (ST-PMA). The results were analysed by multivariate analysis of variance (MANOVA) with three factors (ST-PMA imaging capacity, 2 levels; VVIQ imaging capacity, 2 levels; and mnemotechnic method, 4 levels) and dependent variables immediate recall and delayed recall. All three factors influenced recall. Subsequent univariate analyses of variance indicated that subjects with high ST-PMA score and subjects with high VVIQ score showed better immediate and delayed recall than subjects with low ST-PMA score and subjects with low VVIQ score. Mnemotechnic method (i.e. whether keywords are generated by the experimenter or by peers) significantly affected immediate recall but not delayed recall.