One hundred and twenty-eight kindergarten children were trained on an identity or an oddity task with color and form dimensions to 6/6 or 6/6+30 correct responses and then tested on the transfer task with the same rule-different dimension (S-D task) or the task with the different rule-same dimension (D-S task). When the subjects were trained to 6/6 correct responses and on the oddity rule, the D-S transfer task was learned faster than the S-D task. When the subjects were trained to 6/6+30 correct responses and on the identity rule, the two tasks were learned almost at the same rate. The findings were explained by taking account of the strength of rule and dimensional responses learned in the original task and of the property of the rule learned, and were discussed in relation to the rule and the attention models of relational learning.Key words: rule (identity and oddity) learning, dimensional response, transfer, rule model, attention model.In an extensive review of literature on identity-difference discrimination learning, House, Brown, and Scott (1974) proposed a chaining attentional model. According to this model, oddity learning is represented as acquisition on a chain of three components:(a) attention to a relevant vehicle dimension which carries the oddity relation, (h) attention to the oddity relation among the cues on the relevant dimension, and (c) the instrumental response to the odd stimulus. An important assumption in this model is that for the solution of oddity tasks attention to the oddity relation must be preceded by attention to the relevant dimension. They suggested that this model could be applicable to other types of identity-difference learning with only minor variations.In an earlier study Brown (1970) found that second grade children who were overtrained on an original task learned very rapidly both an intradimensional and an extradimensional oddity tasks. On the basis of this finding Brown suggested the possibility of reversal in the sequential order of the two attentional (or observing) responses:" Once the relational observing response (suggested by House) reached asymptote, the sequential order of the two observing responses would be reversed and the problem would become one of the searching for the oddity relationship (e.g., " find the one that's odd ") , followed by a search for the dimension carrying the odd cue (p. 317)."Recently Bowers (1976) proposed a model which is referred to as a rule model by generalizing the Brown' s suggestion.In this model it is assumed that subjects first select a general problem-solving rule, then attend to a vehicle dimension, and finally choose a cue.Although different terminologies (rule and relation) are used, the two components assumed in the rule and the attention models have very similar implications. However, there is an important difference in the order that the two components are processed. The rule model assumes that subjects select a rule first and then attend to a vehicle dimension, whereas the attention model assumes that subjects attend t...