A handheld and portable device for detecting meat quality parameters was designed, which was composed of hardware components including detecting probe module with multispectral light sources array, constant flow driver module, multispectral data acquisition and processing module, wireless transmission and display module, as well as application software based on Android system on terminal display platform. In addition, multispectral array contained 64 LED light sources which related to the quality parameters and it were divided into eight groups and each group incorporated eight different LED light sources of which the center wavelengths were 475, 515, 525, 575, 610, 760, 810, 910nm respectively, forming a diameter of 5cm brightness uniformity detection range for obtaining spectral data of meat samples. The dimension of this device was 175mm(L)×115mm(W)×25mm(H) and the weight was about 0.45kg. In addition, cost of device was less 2500 RMB. When using multispectral light array to detect samples, some lights were absorbed by samples and another part was reflected from samples which was called diffuse light. After obtaining diffuse light from samples, a silicon photodiode detector with spectral response range of 400~1100nm was installed to receive diffuse light from pork meat in detection zone, and then the signal from silicon photodiode detector were amplified and processed by amplifier chip and A/D converter chip, then different wavelengths spectral data were calculated for establishing meat quality prediction model. For verifying this device, 43 pork samples with different quality attributes were collected for data acquisition and three algorithms including multiple linear regression (MLR), partial least square regression (PLSR) and multiple linear regression (MLR) mathematical with stepwise method were employed to establish pork prediction models. The parameters included color parameters (L*, a*, b*) and total volatile basic nitrogen (TVB-N) content in meat respectively. The 43 samples were divided into calibration and validation sets according to the proportion of 3〖DK〗∶1 to achieve more reasonable prediction results. The results showed that the multiple linear regression (MLR) mathematical with stepwise method had the best prediction model, the correlation coefficients of prediction (Rp) for pork color (L*, a*, b*) and TVB-N content were 0.9471, 0.8504, 0.8563, 0.8027, respectively. At last, the same number of 12 pork samples from two groups of A, B at different time periods were used to verify the feasibility of this model, and the coefficient between predicted values of pork color (L*, a*, b*) and TVB-N content and their true values in each group were more than 0.80. In addition, coefficient of variation (CV) of verifying results in two groups was less than 1.1%, which indicated that this model of this device had certain stability to some extent. This experiment demonstrated that it has the potential in nondestructive detection for assessing meat freshness using this device, which not only meet the requirements of nondestructive testing but also can be widely applied for assessing meat quality in future.