Oceans cover 70% of Earth’s surface but little is known about their waters. While the echosounders, often used for exploration of our oceans, have developed at a tremendous rate since the WWII, the methods used to analyse and interpret the data still remain the same. These methods are inefficient, time consuming, and often costly in dealing with the large data that modern echosounders produce. This PhD project will examine the complexity of the de facto seabed mapping technique by exploring and analysing acoustic data with a combination of data mining and visual analytic methods. First we test the redundancy issues in multibeam echosounder (MBES) data by using the component plane visualisation of a Self Organising Map (SOM). A total of 16 visual groups were identified among the 132 statistical data descriptors. The optimised MBES dataset had 35 attributes from 16 visual groups and represented a 73% reduction in data dimensionality. A combined Principal Component Analysis (PCA) + k-means was used to cluster both the datasets. The cluster results were visually compared as well as internally validated using four different internal validation methods. Next we tested two novel approaches in singlebeam echosounder (SBES) data processing and clustering – using visual exploration for outlier detection and direct clustering of time series echo returns. Visual exploration identified further outliers the automatic procedure was not able to find. The SBES data were then clustered directly. The internal validation indices suggested the optimal number of clusters to be three. This is consistent with the assumption that the SBES time series represented the subsurface classes of the seabed. Next the SBES data were joined with the corresponding MBES data based on identification of the closest locations between MBES and SBES. Two algorithms, PCA + k-means and fuzzy c-means were tested and results visualised. From visual comparison, the cluster boundary appeared to have better definitions when compared to the clustered MBES data only. The results seem to indicate that adding SBES did in fact improve the boundary definitions. Next the cluster results from the analysis chapters were validated against ground truth data using a confusion matrix and kappa coefficients. For MBES, the classes derived from optimised data yielded better accuracy compared to that of the original data. For SBES, direct clustering was able to provide a relatively reliable overview of the underlying classes in survey area. The combined MBES + SBES data provided by far the best accuracy for mapping with almost a 10% increase in overall accuracy compared to that of the original MBES data. The results proved to be promising in optimising the acoustic data and improving the quality of seabed mapping. Furthermore, these approaches have the potential of significant time and cost saving in the seabed mapping process. Finally some future directions are recommended for the findings of this research project with the consideration that this could contribute to further development of seabed mapping problems at mapping agencies worldwide.