While it is well recognised that increasing vegetation cover reduces the Urban Heat Island (UHI) effect in cities, less is understood about the spatial pattern of vegetation type required to maximise cooling benefits. This study examines how different urban vegetation spatial configuration and composition impact on the UHI phenomenon. We investigated this on a set of sites in Metropolitan Melbourne, Australia. Urban vegetation raster data at 20 cm resolution was used to define five height-based vegetation cover types (grass, shrub, small, medium and large trees); and to calculate eight landscape metrics: percentage of landscape (PLAND), mean patch area (AREA_MN), patch density (PD), edge density (ED), mean patch shape index (SHAPE_MN), mean Euclidean nearest-neighbour distance (ENN_MN), Shannon’s diversity index (SHDI) and Landscape shape index (LSI). These vegetation landscape metrics of vegetation types were established as independent variables and statistically analysed with the UHI intensity as the dependent variable using Classification and Regression Tree analysis (CRT). The CRT model was developed based on 6469 records and including depth of 5, 41 nodes and 21 terminal nodes. From the total 43 independent variables, 11 were identified as high impact factors on UHI intensity. Our findings revealed a consistent negative statistical relationship between UHI intensity and PLAND-landscape, PLAND-large tree and PLAND-medium tree across the study sites. The PLAND, ED and AREA_MN were the most prevalent metrics to explain UHI effect, which has also principally demonstrated the joint impact of the three metrics on UHI effect. This study presents a new framework of a fine-scale assessment and modelling for the impact of urban vegetation on UHI; and elaborates a practical approach of using CRT technique to design local-based UHI mitigating strategies taking advantage of different vegetation structure.