The partial circulant diallel cross mating scheme of Kempthorne and Curnow (Biometrics 17: 229-250, 1961) was adapted for the evaluation of genotypes in crosses at the interpopulation level. Considering a random sample of n lines from base population I, and that each line is crossed with s lines from opposite population II, there will be ns sampled crosses that are evaluated experimentally. The means of the ns sampled crosses and the remaining n(n - s) crosses can be predicted by the reduced model <img src="http:/img/fbpe/gmb/v22n2/0291e1.gif" align="absmiddle"> where Yij is the mean of the cross between line i (i = 1,2,...,n) of population I and line j (j = 1',2',...,n') of population II; µ is the general mean, and gi and gj refer to general combining ability effects of lines from populations I and II, respectively. Specific combining ability (Sij) is estimated by the difference <img src="http:/img/fbpe/gmb/v22n2/0291e2.gif" align="absmiddle">. The sequence of crosses for each line (i) is [i x j], [i x (j + 1)], [i x (j + 2)], ..., [i x (j + s -1)], starting with i = j = 1 for convenience. Any j + s -1 > n is reduced by subtracting n. A prediction procedure is suggested by changing gi and gj by the contrasts <img src="http:/img/fbpe/gmb/v22n2/s_ti.gif" alt="s_ti.gif (188 bytes)"> i = <img src="http:/img/fbpe/gmb/v22n2/s_ytra.gif" alt="s_ytra.gif (986 bytes)"> i. - <img src="http:/img/fbpe/gmb/v22n2/s_ytra.gif" align="absmiddle"> .. and <img src="http:/img/fbpe/gmb/v22n2/s_ti.gif" align="absmiddle">j = <img src="http:/img/fbpe/gmb/v22n2/s_ytra.gif" align="absmiddle">.j - <img src="http:/img/fbpe/gmb/v22n2/s_ytra.gif" align="absmiddle">..; the correlation coefficient (r) was used to compare the efficiency of <img src="http:/img/fbpe/gmb/v22n2/s_gtil.gif" align="absmiddle">'s and <img src="http:/img/fbpe/gmb/v22n2/s_tiac.gif" align="absmiddle">'s for selection of lines and crosses. The analysis of variance is performed with the complete model Yij = µ + gi + gj + sij + <img src="http:/img/fbpe/gmb/v22n2/s_e_.gif" align="absmiddle">ij, and the sum of squares due to general combining ability is considered for each population separately. An alternative analysis of variance is proposed for estimation of the variance components at the interpopulation level. An analysis of ear length of maize in a partial diallel cross with n = 10 and s = 3 was used for illustration. For the 30 interpopulation crosses analyzed the coefficient of determination (R2), involving observed and estimated hybrid means, was high for the reduced (g) model [R2 (<img src="http:/img/fbpe/gmb/v22n2/s_yac.gif" align="absmiddle">ij, Yij) = 0.960] and smaller for the simplified (<img src="http:/img/fbpe/gmb/v22n2/s_ti.gif" align="absmiddle">) model [R2 (<img src="http:/img/fbpe/gmb/v22n2/s_yac.gif" align="absmiddle">ij, Yij) = 0.889]. Results indicated that the proposed procedure may furnish reliable estimates of means of hybrids not available in the partial diallel.
