Land-use pattern is spatially correlated and scale-dependent, so doing research on land-use pattern based on raster format data requires making clear how to scale the data and which grain domain is suitable for scaling. The major objective of this study was to explore whether the landscape metrics and spatial autocorrelation index-Moran's I can detect the grain effect of land-use spatial pattern. At first three kinds of scaling methods were carried on the original data to get multi-resolutions images. After comparing several statistics, the better results were interpreted to multi-grains land-use maps. Twenty-three landscape metrics and Moran's I were performed on these maps. Only a few indices were chosen to help delimiting the appropriate grain domain in which the majority of grain-sensitive indices were stable and can be extrapolated or interpolated across spatial grains. The results showed that landscape metrics were grain dependent and could be categorized into three types: regular changing type with obvious inflexions-AI, FRAC_AM, LPI, SPLIT, DIVISION, SHDI, SIDI, MSIDI, SHEI, SIEI and MSIEI; regular changing type without obvious inflexions-NP, PD, LSI, PARA_AM, PARA_MN, PLADJ; unpredictable changing or no changing type-TA, PAFBAC, CONTAG, SHAPE_AM, SHAPE_MN and FRAC_MN. Only the first type was suitable to detect the grain effects of the land-use spatial pattern. Then correlation analysis was performed on these metrics and FRAC_AM, DIVISION and SHDI were picked out as representatives to decide the appropriate grain domain cooperated with Moran's I. This study highlights the need for multi-grain analysis in order to adequately characterize and monitor land-use spatial pattern characteristics, and provides insights into the scaling of land-use spatial pattern.