We propose two novel approaches for the removal of bubble oscillations from marine seismograms. Both methods involve filters containing matched as well as least-squares inverse components. The techniques are deterministic rather than statistical because we require explicit knowledge of the source pulse shape such as a far-field signature. The first approach leads to the design of a matched filter in cascade with a two sided Wiener-shaping operator. The second begins with the computation of the zero-delay Wiener filter that is inverse to the source signature and continues with the autocorrelation of the resulting output.This performance is illustrated with seismic field data. Our techniques are general ones, applicable in principle to any type of source; however, the examples displayed use Maxipulse and Aquapulse seismic records. Acoustic oscillations caused by successive expansions and contractions of gas bubbles initiated by the seismic source can be attenuated effectively with these filters, and the process has been called debubbling by other investigators. Our procedures improve the definition of seismic reflections (resolution), and the debubbling procedure is a first step in data processing prior to applying other deconvolution and pulse compression techniques.