Prior studies analyzed the complexity of the Latin Square task solely based on the relational complexity theory, which, with the relational complexity defined by the complexity of relations processed in paral-lel, could not fully predict children’s performance on the task. So we developed an alternative method to analyze the task complexity by the relational complexity and the necessary processing steps to find a solution. The present study tested the validity of the new method applying to the Latin Square task and investigated how the task complexity influenced children’s performance on the task. 195 pupils from Grade 2―5 were recruited to perform computerized Latin Square task of 15 items. The results showed that: (i) The children’s performance on the Latin Square task fitted perfectly to the predictions by the Rasch measurement model. The relational complexity and the necessary processing steps both validly predicted the children’s reaction time for correct answers and the item difficulty levels assessed by the Rasch analysis. This validated our method for task complexity analysis. (ii) Generally, all the 2nd―5th graders performed well on the items whose relational complexity was binary or ternary. However, they had difficulties in solving the quaternary items, although there was improvement from grade 2 to grade 5.
