High-resolution seismic experiments, employing arrays of closely spaced, four-component ocean-bottom seismic recorders, were conducted at a site off western Svalbard and a site on the northern margin of the Storegga slide, off Norway to investigate how well seismic data can be used to determine the concentration of methane hydrate beneath the seabed. Data from P-waves and from S-waves generated by P–S conversion on reflection were inverted for P- and S-wave velocity (Vp and Vs), using 3D travel-time tomography, 2D ray-tracing inversion and 1D waveform inversion. At the NW Svalbard site, positive Vp anomalies above a sea-bottom-simulating reflector (BSR) indicate the presence of gas hydrate. A zone containing free gas up to 150-m thick, lying immediately beneath the BSR, is indicated by a large reduction in Vp without significant reduction in Vs. At the Storegga site, the lateral and vertical variation in Vp and Vs and the variation in amplitude and polarity of reflectors indicate a heterogeneous distribution of hydrate that is related to a stratigraphically mediated distribution of free gas beneath the BSR. Derivation of hydrate content from Vp and Vs was evaluated, using different models for how hydrate affects the seismic properties of the sediment host and different approaches for estimating the background-velocity of the sediment host. The error in the average Vp of an interval of 20-m thickness is about 2.5%, at 95% confidence, and yields a resolution of hydrate concentration of about 3%, if hydrate forms a connected framework, or about 7%, if it is both pore-filling and framework-forming. At NW Svalbard, in a zone about 90-m thick above the BSR, a Biot-theory-based method predicts hydrate concentrations of up to 11% of pore space, and an effective-medium-based method predicts concentrations of up to 6%, if hydrate forms a connected framework, or 12%, if hydrate is both pore-filling and framework-forming. At Storegga, hydrate concentrations of up to 10% or 20% were predicted, depending on the hydrate model, in a zone about 120-m thick above a BSR. With seismic techniques alone, we can only estimate with any confidence the average hydrate content of broad intervals containing more than one layer, not only because of the uncertainty in the layer-by-layer variation in lithology, but also because of the negative correlation in the errors of estimation of velocity between adjacent layers. In this investigation, an interval of about 20-m thickness (equivalent to between 2 and 5 layers in the model used for waveform inversion) was the smallest within which one could sensibly estimate the hydrate content. If lithological layering much thinner than 20-m thickness controls hydrate content, then hydrate concentrations within layers could significantly exceed or fall below the average values derived from seismic data.