The biologically effective dose (BED) for a non-uniformly irradiated treatment target can be determined using two approaches: averaging BEDs for different voxels in the target and averaging the probability of tumor cell survival in the target. This study compared these approaches for the computation of BED by examining their general properties and comparing BEDs for different treatment plans. The mean BED (BED_mean) determined by averaging BEDs in different parts of the target and BED (BED_nud) determined by averaging the probability of survival were computed for 51 cases of stereotactic body radiation therapy. For each case, two treatment plans were created using dynamic conformal arcs (DCAs) and volumetric modulated arc therapy (VMAT). For the plans with DCAs, the sample mean of BED_mean was 18% > that of BED_nud (for α = 0.35 Gy⁻¹ and α/β =10 Gy). Conversely, for the VMAT plans and the same values of radiobiological parameters, the sample mean of BED_mean was only 5%> that of BED_nud. Theoretically, BED_mean is always ≥BED_nud. When the mean dose in the target is fixed, BED_mean increases with increasing variance of the target dose (σ²). In contrast, BED_nud decreased with increasing σ². Although BED_nud depends on both alpha and beta parameters for the irradiated malignant cells, BED_mean depends only on the alpha/beta ratio. Thus, the use of BED_mean instead of BED_nud can cause incorrect ranking of treatment plans.