We compute the acceleration on the Local Group using 11206 IRAS galaxies from the recently completed all-sky PSCz redshift survey. Measuring the acceleration vector in redshift space generates systematic uncertainties due to the redshift space distortions in the density field. We therefore assign galaxies to their real space positions by adopting a non-parametric model for the velocity field that solely relies on the linear gravitational instability and linear biasing hypotheses. Remaining systematic contributions to the measured acceleration vector are corrected for by using PSCz mock catalogues from N-body experiments. The resulting acceleration vector points approx. 15 degrees away from the CMB dipole apex, with a remarkable alignment between small and large scale contributions. A considerable fraction of the measured acceleration is generated within 40 h-1 Mpc with a non-negligible contribution from scales between 90 and 140 h-1 Mpc after which the acceleration amplitude seems to have converged. The local group acceleration from PSCz appears to be consistent with the one determined from the IRAS 1.2 Jy galaxy catalogue once the different contributions from shot noise have been taken into account. The results are consistent with the gravitational instability hypothesis and do not indicate any strong deviations from the linear biasing relation on large scales. A maximum-likelihood analysis of the comulative PSCz dipole is performed within a radius of 150 h-1 Mpc in which we account for nonlinear effects, shot noise and finite sample size. We obtain β = 0.70 (+ 0.35)(-0.2) at 1 σ confidence level. ; Comment: replace earlier version, references added, errorbars corrected