We present a statistical analysis of voids in the two-degree Field Galaxy Redshift Survey (2dFGRS). In order to detect the voids, we have developed two robust algorithms. We define voids as non-overlapping maximal spheres empty of haloes or galaxies with mass or luminosity above a given value. We search for voids in cosmological N-Body simulations to test the performance of our void finders. We obtain and analyse the void statistics for several volume-limited samples for the North Galactic Pole (NGP) and the South Galactic Pole (SGP) constructed from the 2dFGRS full data release. We find that the results obtained from the NGP and the SGP are statistically compatible. From the results of several statistical tests we conclude that voids are essentially uncorrelated, with at most a mild anticorrelation and that at the 99.5 per cent confidence level there is a dependence of the void number density on redshift. We develop a technique to correct the distortion caused by the fact that we use the redshift as the radial coordinate. We calibrate this technique with mock catalogues and find that the correction might be of some relevance to carry out accurate inferences from void statistics. We study the statistics of the galaxies inside nine nearby voids. We find that galaxies in voids are not randomly distributed: they form structures like filaments. We also obtain the galaxy number density profile in voids. This profile follow a similar but steeper trend to that followed by haloes in voids.