Grain size effect is a major issue in landscape ecology research. Its importance is determined by the effective and precise transformation of the information and characteristics of a landscape pattern as well as the ecological process embodied during the scaling procedure. A landscape pattern that consists of patches has a close relationship with patch classification. In previous studies, grain size effect was analyzed from the perspective of a landscape pattern generated from one particular patch classification. However, differences in the grain size effect caused by different types of patch classification have been ignored. The aim of this study was to explore the effect of different patch classifications on the grain size effect. We used Wuxi City as a study case because it has undergone rapid urban development, has been subject to dramatic changes in land use, and has a vulnerable environment. Three patch classifications were applied: land use/land cover (LULC), urban heat island (UHI), and ecological contribution (EC). Their matching landscape patterns were generated accordingly. In the LULC pattern, the patches were divided into eight categories: cropland, woodland, grassland, garden land, rivers, lakes and ponds, construction land, marshland, and unused land, which were generated from the 2010 Land-use Updating Map for Wuxi. Patches in the UHI pattern were obtained through the following two main steps: (1) Land surface temperatures (LST) were obtained from Landsat TM using the mono-window algorithm. (2)The mean-standard deviation method was employed to transform LST into a thematic map of five thermal categories: very high, high, middle, low, and very low. EC pattern patches were also generated. The ecological system service value (ESSV) of a patch varies depending on the land-use type. The ESSV of patches representing the same land-use type also vary due to its natural features and disturbance from nearby different land-use patch types. Considering its natural features and the received disturbance, the ESSV for each LULC patch was calculated using the multi-weight factors model in the ARCGIS software. The natural breakpoint method was used to transform the LULC pattern into an EC pattern with three value categories: high, middle, and low. The basic spatial unit was 30 m. The pixels scale on the side of the grid cell enabled another 23 basic cells to be assembled, which represented 40, 50, 60, 70, 80, 90, 100, 120, 150, 180, 210, 240, 270, 300, 400, 500, 600, 700, 800, 900, 1000, 2000, 3000 m. Nineteen landscape metrics, including patch density (PD), largest patch index (LPI), landscape shape index (LSI), area-weighted patch fractal dimension (AWMPFD), perimeter area fractal dimension (PAFRAC), cohesion index (COHESION), division index (DIVISION), mean shape index (MSI), and the global spatial autocorrelation index for Moran's I, were computed to detect the LULC, UHI, and EC patterns at different spatial scales using FRAGSTATS and ARCGIS software. The results showed that increasing the spatial grain size changed the response curves for some of the landscape metrics and altered the Moran's I index values. Furthermore, the different patch classifications altered the grain size effect. However, the critical grain sizes for most of the landscape metrics and Moran's I were the same. There was also an interrelationship between grain size effect and patch classification. However, how and to what degree the differences in patch classification alter the grain size effect needs further study.