A method for detecting sub-pixel motion for an optical navigation device is disclosed. The method calculates the displacement of the sub-pixel motion according to brightness curve equations of captured images, which is represented by positional coordinates x and y. The detection method includes capturing a first image frame and a second image frame at different times, selecting plural pixels from the first image frame as calculation reference pixels, generating brightness curve equations of the reference pixels, calculating partial-derivatives of the reference pixel to generate a plurality of two-dimensional first-order equations, and calculating the sub-pixel motion according to the plurality of the two-dimensional first-order equations. The method can reduce the pixel noise by choosing reliable pixels as the reference pixels. The method also reduce the noise by classifying the reference pixels.