Visible and near infrared reflection spectroscopy (Vis-NIRS) was studied as a fast and promising techniques for weed detection. Two kinds of weeds [threeseeded mercury (Acalypha australis L.) and quatrefoil duckweed (Marsilea quardrifolia L.)] and two kinds of crops [soybean (Glycine mas) and corn (Zea mays)] were investigated. A total 120 samples (30 for each species) of the plant leaves were prepared. The spectra were acquired by an ASD Fieldspec handhel spectrometer. After data pre-processing and principle component analysis on spectra from 400 to 1000 nm, one influential outliner was detected. After full-cross calibration of the 79 samples using partial least squares regression, the remaining 40 samples were predicted by the established model. Correlation coefficient of prediction set was 0.987. 100% recognition rate was obtained based on artificial thresholds. So, Vis-NIR spectroscopy is an available alternative for recognizing crops from weeds.