Objective: Local pulse wave velocity (PWV) can be estimated from the waterhammer equation and is an essential component of wave separation analysis. However, previous studies have demonstrated inaccuracies in the estimations of local PWV due to the presence of reflections. In this study we compared the estimates of local PWV from the PU-loop, ln(D)U-loop, QA-loop and ln(D)P-loop methods along the complete human arterial tree, and analyzed the impact of the estimations on subsequent wave separation analysis. Approach: Estimated values were derived from the numerical outputs (pressure, flow, flow velocity, area and diameter waveforms) of a 1D model of the human circulation, and compared against a reference PWV obtained from the Bramwell-Hill equation in a reference configuration, and in a configuration with lower distensibility representing ageing. Main results: When including all nodes, the overall performance of the methods was poor (correlations and mean differences of R-2 < 0.4 and 3.0 +/- 4.1 m s(-1) for the PU-loop, R-2< 0.07 and -0.7 +/- 2.3 m s(-1) for the ln(D)U-loop, and R-2 < 0.06 and -0.4 +/- 2.3 m s(-1) for the QA-loop). Focusing on specific sites, the In(D)U- and QA-loop methods yielded acceptable results in the thoracic aorta and iliac arteries, while the PU-loop method was acceptable at the aortic arch. The reflection-insensitive ln(D)P-loop method performed well over the complete network (R-2 = 0.9 and 0.3 +/- 0.3 m s(-1)), as did a previously proposed reflection-correction method for most vascular sites. Large errors in PWV estimation are attenuated in subsequent wave separation analysis, but the errors are site-dependent. Significance: We conclude that the performances of the PU-loop, ln(D)U-loop and QA-loop methods are highly site-specific. The results should be interpreted with caution at all times.