Knowing the morphology and sediment composition of the seabed is of high importance for various applications. In this contribution, the repeatability of acoustic seafloor classification (ASC) results obtained from MBES backscatter value is investigated. The unsupervised classification algorithm based on Principal Component Analysis has been applied to the MBES backscatter acquired in the Cleaver Bank, Netherlands Continental Shelf, during five different surveys with two vessels. In general, there is good repeatability between surveys demonstrating the potential of using backscatter for marine environmental monitoring. To increase the discrimination performance the so-called depth residuals can be used. These are derived from the bathymetric measurements and considered to be representative for the sediment roughness. The challenge is that the small-scale depth variations are not solely dependent on the sediment roughness but also on the intrinsic uncertainties inherent to the MBES system. An A-Priori Multibeam Uncertainty Simulation Tool (AMUST) has been developed to predict the depth errors induced by various contributors. Correcting the measured depths for these uncertainties, as predicted by AMUST, theoretically provides information about the actual sediment roughness and this should improve the ASC algorithms. This was first tested on a MBES data set from Shallow Survey Conference Plymouth, 2015. It was shown that for the water depth of 20 m the standard deviation of the depth measurements was in agreement with AMUST predictions indicating a smooth seafloor, however, discrepancies between the predictions and real measurements occurred for the water depth of 8 m which is an indication of roughness or morphological features. This indicates the necessity of knowledge about the uncertainties when the objective is to derive the sediment roughness from MBES measurements.