Leaf area index (LAI) is often used as an indicator of green biomass. In tropical forests it can provide important inputs to ecological and physical models that simulate woody biomass accumulation, water-gas exchanges and carbon sequestration. Commonly accepted ground measurements with the LI-COR LAI-2000 instrument were found very difficult to deploy in mountainous, seasonal tropical forests in northern Thailand. As a result, spatial distribution of LAI cannot be derived from remotely sensed data via empirical regression models. Using hemispherical photography, this study developed a new approach to estimate LAI from remote sensing images. A linear unmixing model in the Modified Soil-Adjusted Vegetation Index (MSAVI) domain was first applied to estimate canopy fractional cover (fc) from the JERS-1 optical images acquired with the Very Near Infrared Radiometer (VNIR) onboard the satellite. Based on hemispherical photographs taken at the study sites, a modified Gaussian regression model was developed to relate LAI to estimated fc, with a squared correlation coefficient (R 2 ) of 0.87. Assuming the 1-m pan-sharpened IKONOS image as ground truth, the VNIR-estimated fc values agreed well with the IKONOS fc in 400 randomly selected locations (R 2 =0.79). The rms errors (RMSE) of the VNIR-estimated fc in different forest types ranged from 13% to 21%. With ground hemispherical photographs taken in the wet season, the RMSE of the LAI-fc model developed in this study was only 0.1 in LAI units, indicating that the LAI-fc regression model could be used in all seasons in the study area.