Detection of an underwater target with active sonar in shallow waters such as the Baltic sea is a big challenge. This since the sound beams from the sonar will be reflected on the surfaces, sea surface and sea bottom, and the water volume itself which generates reverberation. Reverberation which will be reflected back to the receiver, is strong in intensity which give rise to many false targets in terms of classifying a target in a surveillance area. These false targets are unwanted and a real target might benefit from these miss-classifications in terms of remaining undetected. It is especially hard if the signal-to-noise ratio (SNR) is approaching zero, i.e. the target strength and the reverberation strength are equal in magnitude. The classical approach to a target detection problem is to assign a threshold value to the measurement, and the data point exceeding the threshold is classified as a target. This approach does not hold for low levels of SNR, since a threshold would not have a statistical significance and could lead to neglecting important data. Track-before-detect (TrBD) is a proposed method for low-SNR situations which tracks and detects a target based on unthresholded data. TrBD enables tracking and detecting of weak and/or stealthy targets. Due to the issues with target detection in shallow waters, the hypothesis of this thesis is to investigate the possibility to implement TrBD, and evaluate the performance of it, when applied on a low-SNR target. The TrBD is implemented with a particle filter which is a recursive Bayesian solution to the problem of integrated tracking and detection. The reverberation data was generated by filtering white noise with an Autoregressive filter of order 1. The target is assigned to propagate according to a constant velocity state space model. Two types of TrBD algorithms are implemented, one which is trained on the background and one which is not. The untrained TrBD is able to track and detect the target but only for levels of SNR down to 4dB. Lower SNR leads to the algorithm not being able to distinguish the target signal from the reverberation. The trained TrBD on the other hand, is able to perform very well for levels of SNR down to 0dB, it is able to track and detect the target and neglect the reverberation. For trajectories passing through areas with high reverberation, the target was lost for a short period of time until it could be retracked again. Overall, the TrBD was successfully implemented on the self-generated data and has a good performance for various target trajectories.