The work aims to explore the feasibility of predicting firmness of plums by ultraviolet radiation/visible spectroscopy technology combined with chemometrics. The spectra acquisition system was used to collect the average spectral reflectance of “Red” and “Green” plums. The standard normal variation (SNV) was used to preprocess original spectral reflectance. Then the successive projection algorithm (SPA) and competitive adaptive reweighted sampling (CARS) method were used to select characteristic wavelengths of 2 (513.04 nm and 636.72 nm) and 10 (230.01, 244.67, 274.71, 287.66, 290.90, 300.59, 311.78, 423.08, 515.39 and 631.31 nm) from 1024 wavelengths, respectively. An error back propagation (BP) network model was established based on full spectra and selected characteristic wavelengths for predicting the firmness of plums. The characteristic wavelengths extracted by SPA and CARS were used as the input of BP network model, which obviously improved the working efficiency of BP network model, and SPA-BP model had the best ability of predicting firmness of plums (rp=0.695, RMSEP=1.610 kg/cm2). Ultraviolet radiation/visible spectroscopy technology combined with the characteristic wavelength selection methods is effective for rapid nondestructive detection on firmness of plums.