The nondestructive detection on firmness of “Red” and “Green” plums is proposed based on hyperspectral imaging technology combined with error back propagation (BP) network.The hyperspectral imaging system is used to collect the hyperspectral image of plums,and the average spectral reflectance of the region of interest (ROI) is acquired.Then the effectiveness of BP network model using derivative,standard normal variation (SNV) and multi-scatter calibration (MSC) is compared and evaluated.Finally,the characteristic spectrum of firmness of plums are extracted by principal component analysis (PCA).The results show that the preprocessing effect of derivative on spectral reflectivity has the best calibration ability of firmness of plums (RC=0.939,RMSEC=0.153),and the preprocessing effect of SNV on spectral reflectivity has the best prediction ability of firmness of plums (RP=0.723,RMSEP=0.580).And principal components with cumulative contribution rate of 99.99 % are selected as the characteristic spectral data in the sample set by principal component analysis,and the dimensionality reduction of the spectral data is well realized,leading to improving the efficiency of BP network model.This study indicates that hyperspectral imaging technology combined with BP network is effective for detection on firmness of plums.