In this article the accuracy of geo-acoustic and geometric parameter estimates obtained through matched field inversion (MFI) was assessed. Multi-frequency MFI was applied to multi-tone data (200-600 Hz) received at a 2-km source/receiver range. The acoustic source was fixed and the signals were received at a vertical array. Simultaneously with the acoustic transmissions, a CTD (conductivity, temperature and depth)-chain was towed along the acoustic track. A genetic algorithm was used for the global optimization, whereas a normal mode model was applied for the forward acoustic calculations. Acoustic data received at consecutive times were inverted and the stability of the inverted parameters was determined. Also, the parameter estimates were compared with independent measurements, such as multi-channel seismic surveys (for geo-acoustic parameters). The obtained uncertainty in the inversion results was assumed to have two distinct origins. The first origin is the inversion method itself, since each optimization will come up with some solution close to the exact optimum. Parameter coupling and the fact that some parameters hardly influence the acoustic propagation further contribute to this uncertainty. The second is due to oceanographic variability. Both contributions were evaluated through simulation. The contribution of oceanographic variability was evaluated through synthetic inversions that account for the actual sound speed variations as measured by the towed CTD-chain.