To improve and simplify the prediction model of carotenoid content of cucumber leaves, genetic algorithm (GA) combined with Metropolis acceptance criterion of simulated annealing algorithm (SAA) as well as interval partial least square (iPLS) were proposed and used to establish the calibration models of carotenoid content against cucumber leaves spectra. The cucumber leaves spectra data were divided into 40 intervals, among which 7 subsets, i.e. No.3, 4, 14, 18, 21, 32 and 33, were selected by SAA-GA-iPLS. The comparison was made between SAA-GA-iPLS and traditional genetic algorithm interval partial least square (GA-iPLS), and the result of this study shows that SAA-GA-iPLS was better than traditional genetic algorithm interval partial least square (GA-iPLS).