The results of generation mean analysis based on twelve generations (Table 1) revealed that even the 10-parameter model could not be fitted to explain the inheritance confirming the involvement of the more complex interaction or linkage in most of the cases. Both the epistatic effects, digenic as well as trigenic were frequently significant with the highest contribution of epistatic effects involving dominance (h). The absolute totals of the second order interactions (Table 2) were many times higher than the main effects and the first order interactions, indicating greater importance of trigenic interactions in controlling the inheritance of this trait. The greater significance of trigenic interaction is also obvious from the fact that the maximum contributing factors in all the crosses were trigenic effects such as parameter dominance x dominance x dominance (z) in the cases Cocorit 71 x A-90-30-1, HI 8062 x JNK-4W-128 and parameter additive x additive x dominance (x) in Raj 911 x DWL 5002. The first order interactions were also higher than the main effects and were quite important. Bhullar et al. (1980), Srivastava et al. (1980) and Singh (1981) also reported the role of epistasis in controlling of this trait in aestivum wheat. While in durum wheat Gill et al. (1983) observed epistasis for this trait and he further observed that it is dependent upon the environments.

In the crosses Cocorit 71 x A-9-30-1, HI 8062 x JNK-4W-128 in late sown and in Raj 911 x DWL 5002 in both the environments, all the three (h), (l) and (z) parameters were significant and differed in signs, indicating duplicate epistasis in sets of three genes whereas in remaining cases conclusion regarding type of epistasis could not be drawn because either (h) or (l) parameter was non-significant. No information is available in durum wheat but in aestivum wheat Bhullar et al. (1980) and Srivastava et at. (1980) also observed duplicate type of epistasis for this character. The absolute totals of the non-fixable gene effects were many times higher than the fixables, indicating their greater importance in the inheritance of this trait (Table 2). Earlier also importance of non- additive gene effects in durum wheat was reported by Widner and Lebsock (1973), Gupta and Ahmad (1979), Srivastava et al. (1982), Gill et al. (1983) and Patil et al. (1997).

Highly significant heterosis over better parent was observed in all three crosses in both the environments (Table 3). Similarly significant inbreeding depression was also recorded except in the cross Cocorit 71 x A-9-30- 1 under normal sown condition. Negative heterosis over better parent was observed in the cross Raj 911 x DWL 5002 in normal sown condition only. Yadav and Narsinghani (2000) reported negative heterobeltiosis and Chakraborty and Tiwari (1995) also reported significant positive heterobeltiosis for this attribute in bread wheat.

Components of heterosis revealed that additive x additive x dominance (x), dominance x dominance x dominance (z) along with additive x additive (i) in the cross Cocorit 71 x A-9-30-1, contributed significantly towards heterosis in both the sowing conditions. In the cross HI 8062 x JNK-4W-128 additive x additive x additive (w), dominance x dominance x dominance (z) along with additive x additive (i) in normal sown and additive x additive x dominance (x), dominance x dominance x dominance (z) along with additive x additive x additive (w) in late sown condition contributed maximum towards heterosis. In the cross Raj 911 x DWL 5002, components additive x additive x dominance (x), additive x additive (i) and dominance x dominance x dominance (z) in normal sown and additive x additive x dominance (x), dominance x dominance x dominance (z) and additive x dominance x dominance (y) in late sown condition, contributed towards heterosis. In the cross Cocorit 71 x A-9-30-1 under late sown environment inbreeding depression was significant and negative which indicated improvement of F₂ over F₁ mean for grain weight.

This study shows that the epistatic interactions particularly at trigenic level played a greater role in controlling the inheritance of grain weight in all three crosses under both the sowing conditions, the absolute totals of non-fixable effects were higher than the fixable effects. Breeding procedure that may utilize both additive and non-additive gene effects would prove beneficial. The diallel selective mating and biparental mating that may exploit both the types of gene effects could be worthwhile for the genetic improvement of grain weight in these crosses under normal sown environment to enhance grain yield in durum.