We present a robust and automatic method for evaluating the 3-D navigation accuracy in ultrasound (US) based image-guided systems. The method is based on a precisely built and accurately measured phantom with several wire crosses and an automatic 3-D template matching by correlation algorithm. We investigated the accuracy and robustness of the algorithm and also addressed optimization of algorithm parameters. Finally, we applied the method to an extensive data set from an in-house US-based navigation system. To evaluate the algorithm, eight skilled observers identified the same wire crosses manually and the average over all observers constitutes our reference data set. We found no significant differences between the automatic and the manual procedures; the average distance between the point sets for one particular volume (27 point pairs) was 0.27 +/- 0.17 mm. Furthermore, the spread of the automatically determined points compared with the reference set was lower than the spread for any individual operator. This indicates that the automatic algorithm is more accurate than manual determination of the wire-cross locations, in addition to being faster and nonsubjective. In the application example, we used a set of 35 3-D US scans of the phantom under various acquisition configurations. The US frequency was 6.7 MHz and the average target depth was 6 cm. The accuracy, represented by the mean distance between automatically-determined wire-cross locations and physically measured locations, was found to be 1.34 +/- 0.62 mm.