Soybean rust is one of the most destructive foliar diseases of soybean primarily because it produces copious amounts of air-borne spores that can infect large areas of soybean production causing significant yield losses if left unchecked. Timely application of fungicide in the early stage of rust infection is critical for effective control of the disease, and heavily relies on the capability of detecting the degree of infection or severity. This paper reported research outcomes from developing an image processing method for quantitatively detecting rust severity from multispectral images. A simpler and faster threshold tuning method was developed based on HSI (Hue Saturation Intensity) color model for segmenting disease infected area from plant leaves. Two disease diagnostic parameters, i.e. ratio of infected area (RIA) and rust severity index (RSI), were extracted and used as symptom indicators for quantifying rust severity. To realize timely and automatic rust detection, another method of analyzing the centroid of leaf color distribution in the polar coordinate system was investigated to replace the segmentation approach. Plant images with various levels of rust severity were collected to support this research. Test results proved that the segmentation method was capable of detecting degrees of soybean rust severity under laboratory conditions by calculating RIA and RSI. The centroid locating method had a potential to be used for practical application in the field.