The aim was to identify optimized combinations of Streptomyces griseus protease concentration (CONC), incubation length (TIME), or amount of crude protein (CP) incubated in buffered enzymatic solution (CPW) to predict the in vitro rumen-undegraded feed CP (RUP) of 26 different feeds (soybean, rapeseed or sunflower meals, wheat bran, distillers dried grains with solubles, maize co-products and alfalfa hay). Different levels of CONC (0.08, 0.19, 0.44, 0.69 and 0.80 enzymatic units [U] of S. griseus protease/ml), TIME (6, 10, 18, 26 and 30 h) and CPW (69, 118, 235, 353 and 401 mg CP) were tested in agreement with a central composite design (CCD) with four replications of the central point to calculate second-order polynomial equations of main tested effects. The RUP was estimated by incubating samples in a buffered rumen fluid for 16 h or by adopting different enzymatic approaches as planned a priori in CCD. Differences between rumen and enzymatic RUP (ΔRUP) were estimated and regression terms of second-order polynomial equations for estimating ΔRUP were calculated between and within feeds. These equations were optimized using the non-linear generalized reduced gradient method with the objective set at ΔRUP equal to 0. The adoption of CCD permitted identification of optimized enzymatic combinations of CONC (0.12 U of S. griseus protease/ml), TIME (18 h) and CPW (from 233 to 458 mg CP for distillers dried grains with solubles and soft white wheat bran, respectively) to predict RUP accurately in all feed categories except for soybean meal, where optimized combinations were 0.47 U of S. griseus protease/ml, 18 h and 435 mg CP.