Fresh tea leaves are mainly composed of water, total sugar, tea polyphenols, caffeine, and protein. The water (dry matter) content and extract are important indicators to monitor the plant irrigation and freshness of tea, particularly on evaluating tea brewing. Tea polyphenols are one of the most important healthy components in fresh tea leaves. Chemical analysis and sensory evaluation are two traditional ways of tea quality evaluation. The chemical analysis process is cumbersome and time-consuming, while the sensory evaluation is subject to the subjective influence. Both are destructive testing. The Visible-Near-Infrared(VIS-NIR) spectroscopy can characterize the data related to sample character, providing the possibility of non-destructive testing of tea quality. This study aims to explore the rapid detection for muti-quality of fresh tea leaves using the VIS-NIR spectroscopy, and thereby a portable device was developed suitable for the tea leaves. The self-developed portable equipment was composed of the host and handle parts. The host part included a spectrometer, light source, rechargeable battery, voltage regulator board, and cooling fan, with the approximately size of 240 mm×250 mm×240 mm. The size of the handle part was approximately 130 mm×100 mm×30 mm. In the core component of handle part, the gear-rack drive system ensured manually opening the blade clamp by pulling the button, and then automatically reset under the traction of the spring. In addition, the reference boards were designed to collect the black and white reference for the real-time correction of working state in the portable equipment. The raw VIS-NIR diffuses reflectance spectra of tea were collected using the portable device and four preprocessing, including the normalization(NOR), First Derivative(FD), Standard Normal Variable transformation(SNV), and Probabilistic Quotient Normalization(PQNOR), aiming to correct the noise and scattering effects in the raw spectra. The quantitative prediction models of Partial Least Squares (PLS) were established for the dry matter, water extract and tea polyphenol content using the different preprocessing. A best accuracy was achieved in the PLS model using the PQNOR preprocessing spectra. The correlation coefficients in the verification set for dry matter, water extract, and tea polyphenol content were 0.905, 0.896 and 0.747, respectively. The Root Mean Square Errors (RMSE) in the verification set were 0.860%, 0.559% and 0.549%, respectively. Furthermore, the established model was written into the software in the device, where verified in the tea garden. The rest of 20 samples without modeling were used as the prediction set to test the stability and accuracy of the device. The stability of the device was evaluated by the relative range of test datum, and the accuracy was assessed by the RMSE of predicted mean and measured value, where the measurement was repeated ten times for each sample. The test results showed that the repeatability of the device was within 5%, and the RMSE of dry matter, water extract and tea polyphenol content in the prediction set were 0.903%, 0.634% and 0.551%, respectively. In each fresh tea leaf, the detection speed of the device was about 1s. The prediction accuracy met the requirements of on-site use. Fresh tea leaves from 4 tea gardens were pictured. The samples from multiple tea gardens effectively expanded the range of dry matter, water extract, and tea polyphenol content, providing the possibility of establishing a highly adaptable predictive model. Nevertheless, this study was conducted in summer and autumn, where the quality of tea varies quite distinctly in different seasons. In the future, the modeling samples can be extended to cover those from the origins, varieties, and seasons. [ABSTRACT FROM AUTHOR]