Diabetes is a familiar disease in modern society. In the early stage of diabetes, symptoms are unobvious, but they usually induce diabetic autonomic neuropathy or, worse, cardiovascular autonomic neuropathy. Pupillometers are effective instruments for observing human pupils. This article presents a novel wearable pupillometer design, without external light artifacts, and an embedded algorithm with blinking elimination, which investigates autonomic neuropathy through recording pupil dynamics triggered by an external sensitive invisible light source. The pupillometer is experimented on 36 healthy subjects and 10 diabetic patients under four different colors (white, red, green, and blue) as well as two different light intensities: 50 and 500 mcd. Ten parameters derived from pupil diameter, pupil response time, and pupil response speed will be evaluated for the healthy subjects and diabetic patients. The results show that three in four parameters related to pupil diameters, one in four related to light intensities, and one in two related to pupil response speed could have significant differences (p<0.05) between healthy subjects and diabetic patients. These parameters obtain over 85% sensitivity, 83% specificity, and 88% accuracy. The pupillometer is proven reliable, effective, portable, and inexpensive for diagnosing diabetes in an early stage.