This paper aims at reducing the bias distortion in familiar sub-pixel analysis algorithm in miniature spectrometer. The process of sub-pixel analysis is modeled and the fallibility in reported sub-pixel analysis algorithm is discussed. The discussion suggests that an important inducement for distortion is cumulated errors. Hence, an averaged sup-pixel analysis algorithm is put forward. According to the new algorithm, the same iteration is executed twice but in opposite directions and results of the same sub-pixels are averaged. Relative simulation indicates that compare with old ones the new algorithm reduces bias distortion by several times and contributes to an enhanced resolution and wavelength accuracy.