For non-small cell lung cancer (NSCLC), unsatisfactory local control (LC) still remains an important cause of failure. It has been suggested that improved LC can be achieved with both higher radiation dosage and adequate target coverage. Modern three-dimensional treatment planning systems (3D-TPSs) offer many tools for planning optimization. Biophysical models, which estimate the normal tissue complication probability (NTCP), are gaining in importance in comparing plans. This study compares conventional two-dimensional (2D) with 3D irradiation techniques using parameters related to volumetric dose distribution and two different biophysical models predicting normal tissue tolerance to radiotherapy (RT). Nine patients with inoperable locally advanced NSCLC were treated with a beam's eye view-based 3D technique. For the same patients, a conventional treatment was simulated; the irradiation geometry and beam contour were fully defined at the simulator and then transferred to the 3D-TPS to calculate the dose distribution. Both techniques gave the same prescribed dose at the reference point. Dose-volume histograms (DVHs) and dose statistics of organs at risk (OARs) (heart, lung(s), parenchyma lung, spinal cord and oesophagus) were analysed. The probability of side effects was estimated using two different biophysical models: the integrated normal ("empirical") model and the relative seriality model. Apart from contralateral lung, the 3D irradiation technique significantly reduced the average mean doses to all OARs. The current analysis suggests that in the treatment of locally advanced NSCLC, the use of 3D irradiation techniques allows a large sparing of OARs; this advantage is confirmed by both dose statistics analysis and NTCP values.