The effect of instruction on the length-difficulty relationship was studied for a complex rational learning task. Data were obtained for problems 4-, 6-, 8-, and 10-units in length, both with and without instruction in the prinicples underlying solution of the problems. The task involved the discovery of the correct sequence in which to de-press a row of buttons, utilizing the information on relative sequency position which could be deduced from the dependency relations of a corresponding panel of lights. The Ss were 160 basic trainees at Lackland Air Force Base. It was found that when no specific instruction was given, the 6-, 8-, and 10-unit problems were of approx. equal difficulty, while the 4-unit problem was significantly less difficult. With instruction in underlying principles, on the other hand, the 4-, 6-, and 8-unit problems were found to be of approx. equal difficulty, while the 10-unit problem was now relatively more difficult. In addi-tion to the shift in relative difficulty, the detailed instructions were effective in reducing error scores generally for all of the conditions.