Response surface methodology was used to optimize formulations of chocolate peanut spread. Thiq-six formulations with varying levels of peanut (25-90%). chocolate (570%) and sugar (555%) were processed using a three-component constrained simplex lattice design. The processing variable, roast (light, medium, dark) was also included in the design. Response variables, measured with consumers (n = 60) participating in the test, were spreadability, overall acceptability. appearance, color, flavor, sweetness and texture/mouthfeel, using a 9-point hedonic scale. Regression analysis was performed and models were built for each significant (p < 0.01) response variable. Contour plots for each attribute, at each level of roast, were generated and superimposed to determine areas of overlap. Optimum formulations (consumer acceptance rating of 2 6.0 for all attributes) for chocolate peanut spread were all combinations of 29-65% peanut, 9-41 % chocolate, and 17-36% sugar, adding up to loo%, at a medium roast. Verification of two formulations indicated no difference between predicted and observed values.

Response surface methodology was used to profile and characterize formulations of chocolate peanut spread. A constrained mixture design for 36 different formulations with varying peanut (P), chocolate (C) and sugar (S) was used. A processing variable, roast (R), was included where peanuts were roasted to light, medium and dark levels. A descriptive panel (n = 10) identified and rated 24 attributes, using 150‐mm unstructured line scales. Regression analysis was performed and models were reduced. Models having R2 > 0.70 were selected for prediction. Contour maps were constructed to: (1) visualize the effects of mixture components and roasting level and (2) characterize optimum formulations at light, medium and dark, which were determined as (30–49% P, 23–40% C and 21–31% S); (29–65% P, 0.9–41% C and 17–36% S) and (27–56% P, 19–45% C and 18–35% S), respectively, adding up to 100% of the mixture. Analyses of optimum and nonoptimum formulations and significant differences were not found between predicted and observed values for most attributes.