An experiment was designed to compare the accuracy and reproducibility of three different techniques for in vivo magnetic resonance (MR) imaging-based volume measurements of brain structures. These techniques were tracing, thresholding, and random marking. Anterior temporal lobe (ATL) and hippocampal formation (HF) volumes in 10 volunteers were measured from MR images, as were four cylinders of known volume. The upper limit of accuracy of in vivo volume measurements is estimated to be within 0.1 cm3 of true volume for the HF and 0.9 cm3 for the ATL with a combined tracing-thresholding technique. Intra- and interobserver variations were estimated from the pooled standard deviations of HF and ATL measurements. With the combined tracing-thresholding technique, the coefficient of variation for HF measurement was 1.9%; for the ATL measurement, it was 0.7%. The results indicate that MR-based volume measurements of these brain structures can be made with high precision and reproducibility.