Cokriging, a multivariate best unbiased linear estimator, has been widely used in many fields. This paper has applied cokriging methods to remote sensing image fusion. The variograms from low-resolution and high-resolution images and the cross-variogram between the images were calculated, and then theoretical models were formulated for the extraction of structural information from the images. Three cokriging variants were used for image fusion, namely, simple cokriging, ordinary cokriging and standardized ordinary cokriging. The experimental results indicate that the simple cokriging yields the best spatial quality result whereas the ordinary cokriging produces the best spectral quality compared with the other cokriging variants. The cokriging methods were further compared with principal component analysis in terms of fusion quality, and the result shows that the fusion quality of the cokriging methods is fairly satisfactory.