Generation mean analysis was carried out in six generations (P1, P2, F1, F2, BC1 and BC2) of five families involving improved cultivars of yellow sarson. The simple scaling and joint scaling tests indicated the presence of non-allelic interaction and found 6-, 5-, 4- and 3-parameter model adequate for yield and yield contributing characters in different families. Both additive and non-additive gene effect was found important for seed yield per plant and yield contributed characters, which directly associated with yield. In most of the families, dominance x dominance [l] interaction was larger than the additive x additive [i] and additive x dominance [j] effects, whereas for the main effect, the dominance [h] component was greater than the additive [d] component. The dominance [h] and dominance x dominance [l] effects were in opposite direction, suggesting the occurrence of duplicate epistasis and suggested chance of getting transgressive segregants in the succeeding generations. The study indicated that selection and heterosis breeding methods can be used to exploit fixable and non-fixable gene action.