As existing filtering methods could not overcome the contradiction between noise removal and feature preserving for the light-spot signals on a CCD, the random error resulted from image noises has become a key constraint on the location accuracy of the light-spot images. Therefore, a filtering method named Attribute Distance Weighted Average (ADWA) is proposed to denoise the light-spot images. The method separates signals and noises based on their attributes, and alleviates the contradiction between noise removal and feature preserving of the signals progressively through introducing new attributes. Experimental results show that as compared with the bivariate ADWA based on “location” and “value”(2-ADWA/LV),the sub-pixel location accuracy of light-spot has been improved by more than 20% and the standard deviation of location reduced to 0.024 pixel by introducing a 3-ADWA/LVG formed by adding an attribute "gradient" to 2-ADWA/LV. It concludes that when the ADWA is used to denoise the light-spot image, the filtering performance can be further enhanced progressively in both noise removal and feature preserving through introducing new attributes. The ADWA gives a new effective way to further improve the sub-pixel location accuracy of the light-spot.