The paper presents the temporal and spatial differences of island landscape gradient change and coastal landscape fragmentation in Dongshan island. Large patch index (LPI) and patch density (PD) were chosen as landscape fragmentation index. Based on remote sensing images of 1994, 2003, and 2011, Dongshan island was divided into 7 landscape types. First, in the landscape classification maps, 26 buffer zones, each being 500m wide, were divided from the coast to inland, then, the landscape spatial patterns of each buffer zone were analyzed with FRAGSTATS. Second, the characteristic scale of landscape fragmentation was determined by using the semi-variable function and the moving window method. Third, according to the determined characteristic scale, moving window method in FRAGSTATS was carried out to analyze the spatial distribution of LPI and PD. Finally, four sample lines, which extend 200m, 500m, 1000m from coastline and across the center of the town were set, and the sampling points for every 200 m were selected on the sample lines, using the sample command of ArcGIS, then, the LPI and PD value of each sampling point were extracted. The results showed that the landscape fragmentation of the buffer zones increased with time generally. The landscape fragmentation continued to decrease in the coastal zones, the northwest and the south zones with forest, while in the town and farmland zones of the island, the fragmentation were increasing. The characteristic scale for analyzing landscape fragmentation of Dongshan island was 1000m. On this scale, it was found that the change rate of landscape fragmentation indicators had obvious spatial difference. Then,landscape fragmentation characteristics of the extracted sample points showed the coastal zone of 500m and 1000m from coastline were disturbed strongly by human, and the fragmentations were growing. The artificial disturbance, which tended to be scattered, created fluctuation significantly of fragmentation index curve with time. Urbanization, tourism, and agriculture development should be the main driving factors of landscape fragmentation. Using Pearson correlation coefficient, we analyzed the correlation of landscape fragmentation and other landscape pattern, which were found to have strong correlation. It indicated that chosen landscape fragmentation index can reflect the basic characteristics of landscape pattern change of the study area.