Seismic data are usually stored using a 32-bit floating-point format for each data sample after conversion of the field tapes to whatever internal format is used by the processing system. This procedure ensures that there is no possible loss of dynamic range as field data are always recorded with less than 32 significant bits, and also minimizes the amount of work that the central processing unit (CPU) has to do, as data may be transferred without conversion from the mass storage device. However, the increasing power of modern CPUs offers the possibility of using a compact format of less than 32 bits per sample in the mass storage medium, and using the CPU to expand the data to 32-bit floating-point format once it has been read in. This procedure reduces the input/output (I/O) load at the cost of increasing the CPU load, which is not unreasonable as CPU costs are at present decreasing much faster than I/O costs. Compact storage has a number of operational benefits: if more data can be fitted on to each tape, the number of tape changes per shift is reduced proportionately, and smaller tape libraries are required. Disc capacity is almost invariably a scarce resource, and any increase in seismic data storage capacity at zero capital cost will be met with broad smiles from the system manager. A further benefit accrues to processing centres which do not have operators on duty for 24 hours per day to mount and dismount tapes. The amount of work which can be done during the unattended period is often limited by the amount of data which can be stored on disc. Tape drives are often a bottleneck in system productivity and they are seldom used at their maximum speed. If processing can be done from disc, the tapes are used as spooling devices and can operate much closer to their maximum capacity. The actual increase in storage capacity will be smaller for tapes than for disc, as most internal tape formats are unblocked; they have an end-of-record gap consisting of half an inch of blank tape between each data trace. For a high density, 6250-bpi, 9-track tape, the data samples of a 1500-sample trace using 32 bits per sample will only occupy 1 inch of tape, so that each trace requires 1.5 inches of tape. Using a 16-bit format for the data samples would reduce the amount of tape needed to 1 inch, thus raising the number of traces fitting on to a reel by 50%. For disc, the storage capacity would be nearly doubled by using 16-bit data storage.