Tomographic-PIV is a recently developed three-dimensional velocimetry technique relying on the reconstruction of the instantaneous three-dimensional tracer particle distribution from their recorded images by means of a tomographic algorithm. Although previous experiments have shown good potential for the technique returning the expected 3D flow structure, the results so far have never been assessed quantitatively. Therefore the present study proposes an experimental investigation to quantify both 3D particle reconstruction noise and velocity accuracy in Tomographic-PIV. Measurements of a circular cylinder wake flow at ReD= 2700 are carried out varying the most relevant experimental parameters, which are the light sheet thickness (ranging from 4 up to 12 mm) and particle image density (ranging from 0.01 to 0.1 ppp). From the returned reconstruction volumes the noise level (or ghost particle density) is extracted, which results in the reconstruction accuracy as a function of the mentioned experimental parameters. Based on this function the reconstruction accuracy in other experiments can be estimated. Finally the velocity accuracy is determined from a comparison with Stereo-PIV. It is shown that the mean and RMS velocity agree within 0.5 and 0.3 m/s respectively (corresponding to 0.30 and 0.18 voxel particle displacement). Furthermore it was found that reconstruction noise or ghost particles have a limited effect on the crosscorrelation. For the range of particle image seeding densities considered (0.02 to 0.08 ppp) no significant difference in the returned flow statistics was observed, even though the number of ghost particles exceeded the number of actual particle for 0.08 ppp.