A probabilistic model of reflection-impulsivity as measured by the Matching Familiar Figures Test (MFFT) is presented and tested on 77 fourth graders. In testing the model two groups emerged, a random response group (n = 22) and a cognitive response group (n = 55), who use the evaluation process. Correlations among latencies, total errors, initial errors, school-scores, and IQs were compared for our total sample and the two groups. The correlation of MFFT latencies and errors disappeared for the random-response group and new correlations with errors appeared while the cognitive response group kept a statistically significant correlation of latencies with errors and no other error correlation was shown. This last group was classified by Kagan's median-split procedure showing that most reflective subjects kept their classification. Preliminary results suggest a reinterpretation of reflection-impulsivity literature, especially as related to the impulsive style.