Vegetation cover maps of three different years (e.g. 1989, 1997 and 2003) were derived from the Landsat TM images for the city of Shanghai. Eight types of cover change trajectories were then identified from the maps, and their spatial patterns were analyzed from the perspective of landscape ecology. The Logistic regression was used to model the relationship between the vegetation change trajectories and 12 spatial, demographic, and landscape factors. The analytic results led to three most important conclusions. (1) Within the concerned time period, the total vegetated area in Shanghai followed a continual downward trend, particularly in the Pudong new district. Trajectories of remaining vegetated in all three years comprised more than 50% of the total city area, and trajectories of non-vegetated throughout the entire period made up about 20%. Transformation from vegetation to non-vegetation in the first phase (1989 to 1997) took place largely in the surrounding area of the urban districts, whereas the same process in the second phase (1997 to 2003) often occurred farther away from the urbanized area. The fact that some trajectories exhibited an alternation between vegetation and non-vegetation suggested the existence of a process of vegetation restoration instead of permanent destruction or removal of vegetation over time. (2) The results from the Logistic regression model indicated that the distance from roads had the strongest influence on vegetation change trajectories. The rest of the factors can be ranked in descending order by the level of influence as follows: distance from district centers, distance from the patch border between vegetation and non-vegetation (1997), distance from edge between vegetation patch and non-vegetation (1989), distance from business center, population density (1990), distance from river, distance from expressway, land use diversity within 100 meters, population density difference between 2003 and 1990 and distance from the center point of Shanghai. (3) The Logistic regression models produced a satisfactory accuracy; however, the binary logistic model performed better than the multinomial logistic model being used for this study. Overall speaking, the factor of distance from roads seemed to have a strong influence on vegetation changes. The alternating vegetation change trajectories in this study demonstrated a rather complicated changing pattern in parts of Shanghai′s vegetation cover from 1989 to 2003, implying the existence of spatial dependence and temporal dependence in the changing process.