The need for high-throughput assays in molecular biology places increasing requirements on the applied signal processing and modelling methods. In order to be able to extract useful information from the measurements, the removal of undesirable signal characteristics such as random noise is required. This can be done in a quite elegant and efficient way by the minimum description length (MDL) principle, which treats and separates 'noise' from the useful information as that part in the data that cannot be compressed. In its current form the MDL denoising method assumes the Gaussian noise model but does not require any ad hoc parameter settings. It provides a basis for high-speed automated processing systems without requiring continual user interventions to validate the results as in the conventional signal processing methods. Our analysis of the denoising problem in mass spectrometry, capillary electrophoresis genotyping, and sequencing signals suggests that the MDL denoising method produces robust and intuitively appealing results sometimes even in situations where competing approaches perform poorly.