Eleven cultivars and breeding lines were chosen for use as parents of a half-diallel cross. Nine of the parents were selected on the basis of BLUPs of breeding value for the roasted peanut, sweet, and bitter sensory attributes (Pattee et al., 2001, 2002a); they included two runner-type lines [F439-2-3-2-1, a sibling of Florunner (Norden et al., 1969), and Andru 93 (Gorbet and Knauft, 1995)], four virginia-types [Florida breeding line UF714021, a component of the Altika cultivar (Norden and Gorbet, 1974), NCSU breeding lines N97054J selected from cross X90053 (Pattee et al., 2002b) and NC Ac 18457, and landrace selection White's Runner], two spanish-type cultivars [Pearl and Pronto (Banks and Kirby, 1983)], and one valencia-type cultivar [New Mexico Valencia C (Hsi, 1980)]. In addition to the parents predicted to have positive effects on peanut flavor, two additional parents were included because of their widespread use among peanut growers: runner-type cultivar Georgia Green (Branch, 1996) and virginia-type cultivar Perry ( Isleib et al., 2003). Georgia Green was later found to have positive BLUPs of breeding value for flavor (Pattee et al., 2003) while Perry was a cultivar with negative BLUPs (Pattee et al., 2001, 2002a). The 11 parents were crossed in a half-diallel mating in the summer of 2001 in the greenhouse at the NCSU campus in Raleigh, NC. Each parent was used five times as a male and five as a female in the mating design.

F₁ hybrids were grown in a winter nursery at Juana Diaz, PR, in the winter of 2001–2002. F₂ populations were grown at the Peanut Belt Research Station (PBRS) in Lewiston, NC, in 2002, and single plant selections were made on the basis of pod characteristics and maturity as evidenced by bright hull color after the inverted peanuts had been allowed to dry in the windrow for several days. Selections were made in the F₃ at PBRS in 2003. F_3∶4 families from the same cross were bulked in equal numbers to provide seed for testing in a replicated trial at PBRS in 2004. Thirty-three crosses survived the inbreeding and selection program; these were grown with check cultivars Florunner, NC 7 (Wynne et al., 1979) and Gregory (Isleib et al., 1999) in a randomized complete block design with two replications. Each plot comprised two rows 7.3 m in length spaced 91 cm apart. Seed spacing at planting was 25 cm. Plots were planted on 11 May 2004, dug 1 October 2004, and harvested by combine on 7 October 2004.

Pods were sized on a rolling grader and shelled with a standard reciprocating grading sheller (Georgia Federal-State Inspection Service, Albany, GA). Seeds were screened over a 6.0 × 25.4 mm (15/64 × 1 in) grading screen. The shelled peanuts were stored at 5 C and 60% RH. Samples were roasted in May using a Blue M “Power-O-Matic 60” laboratory oven, ground into a paste, and stored in glass jars at −20 C until evaluated. The roasting, grinding, and color measurement protocols were as described by Pattee and Giesbrecht (1990). Roast color of the peanut paste was evaluated as CIELAB L*.

An eight-member, trained, roasted peanut profile panel at the Food Science Dept. at NCSU, Raleigh, NC, conducted a descriptive sensory analysis of all peanut-paste samples using a 14-point intensity scale. Panel orientation and reference control were as described by Pattee and Giesbrecht (1990) and Pattee et al. (1993). Sensory evaluation commenced on 7 October 2005 and continued until all samples were evaluated. The averages of individual panelists' scores on sensory attributes were used in all analyses in this study.

Data from the field experiment were analyzed using the general linear models procedure (PROC GLM) of the SAS statistical software package (SAS Institute, Cary, NC). A design matrix was developed to partition the total variation among the 36 entries in the trial into parts due to (1) the difference between the three checks and the hybrid populations, (2) variation among the three checks with one contrast between Florunner and the two virginia checks and a second between NC 7 and Gregory, and (3) variation among the hybrid populations based on Griffing's (1956) Model 1 Method 1. All 11 parents were represented in the hybrid populations, and general combining ability (GCA) was estimable for all 11. Because only 33 of the 55 crosses survived the inbreeding and selection phase of population development, not all specific combining ability (SCA) effects were estimable. Those effects that were not estimable were assumed to be zero, and the remaining SCA effects were left in the model. When variation due to SCA was not significant, the model was run without SCA effects to allow estimation of GCA effects without the distortion that arises from confounding of SCA with GCA effects in the unbalanced set of crosses. GCA estimates were separated by t-test. The need for use of the covariates fruity and roast color (linear and quadratic) to control error variation was tested for each response variable. Covariates were retained in the model if their effects were significant. If not, they were removed from the model. The error mean square from the full model was used in all tests of significance and in calculating standard errors of effects.

Mean values for intensities of the roasted peanut, sweet, and bitter attributes for the 11 parents were calculated from the same set of data used to calculate BLUPs of breeding value (Pattee et al., 2003). Means were adjusted to common year and location effects and also for the effects of the covariates roast color (linear and quadratic) as an indicator of degree of roast and intensity of the fruity attribute, an indicator of poor environmental conditions pre- and post-harvest. Simple correlations were calculated between GCAs and BLUPs and between GCAs and means.