The Stroop-phenomenon in its single-trial reaction-time version is analyzed in terms of Donders's additive-components approach. The complete set of component tasks forms an incomplete factorial design, and so, for resolving reaction times into their component parts multiple regression with dummy coding is used. The results of two independent experiments confirm the asymmetry of the incongruence effect in naming and reading, but contradict the redundancy hypothesis for congruent items. This finding is consistent with the linear approach which in Exp. 1 accounts for about 45%, and in Exp. 2 for about 30%, of total reaction-time variance. In a third experiment reaction times for the detection of colours/words and for congruence/incongruence are measured. These data confirm the estimation of both the basic constant and the assumed matching component in the first two experiments. Finally it is proposed that the data can be interpreted on the assumption that a matching process is responsible for the incongruence effect in naming. Reading interference does not normally occur because of faster processes or fewer stages in a parallel 'reading' channel. Different strategies in coping with the Strooptask ('reading' vs 'inhibition') may explain differences between Exps. 1 and 2 and the absence of the redundancy effect from the data. The constraints on such assumptions in the context of reaction-time measurement are discussed. © 1979 Springer-Verlag.
