In this paper we present a data compression scheme that is especially suited for the compression of data sets resulting from computational fluid dynamics (CFD). By adopting the concept of the JPEG image compression standard and extending the approach of Schmalzl (2003), we employ a three-dimensional discrete cosine transform of the data. The resulting frequency components are rearranged, quantized, and, finally, stored using standard variable length integer codes. The compression ratio and also the introduced loss of accuracy can be adjusted over a wide range by means of two compression parameters to give the desired compression profile. Using the proposed technique, compression ratios of more than 60:1 are possible with a mean error of the compressed data of less than 0.05%.