For short range high linear energy transfer (LET) radiation therapy the biological effects are strongly affected by the heterogeneity of the specific energy (z) distribution delivered to tumour cells. Three-dimensional (3-D) dosimetry information at the cellular level is required for this study. An ideal approach would be the reconstruction of the cell and the radiation source microdistribution from sequential autoradiographic sections, which is, however, not a practical solution. In this paper, a novel microdosimetry analysis method, which obtains the specific energy (z) distribution directly from the morphological information in individual autoradiographic sections, is applied to human glioblastoma multifore (GBM) and normal brain tissue specimens in boron neutron capture therapy. The results are consistent with Monte Carlo simulation and demonstrate a uniform radiation source distribution in both GBM and normal brain tissues. We also hypothesise a biophysical model based on specific energy for survival analysis. The specific energy distributions to cell nuclei were calculated with a uniform radiation source distribution. By combining this microdosimetric analysis with measured cell survival data at the low dose region, a cell survival curve at high doses is predicted, which is consistent with the commonly used simple exponential curve model for high LET radiation.