Abstract: In many fields, there are two types of data: hard data and soft data. Soft data typically provide an extensive coverage of the field under study although with low resolution. It is necessary to condition the reconstructed models to all these different types of data so as to improve the accuracy. The statistical information reconstruction of images will be difficult and inaccurate when only hard data are available or there are no conditional data. Accuracy of reconstructed images can be improved through the use of soft data during the process of reconstruction. Integrating soft data with hard data, a method based on MPS (multiple-point geostatistics) is proposed to reconstruct statistical information of images. By reproducing high-order statistics, MPS allows capturing structures from a training image, and then anchoring them to the specific model data. A training image is a numerical prior model which contains the structures and relationship existing in realistic models. During the process of regenerating characteristic patterns in a training image, the accuracy of reconstructed images is improved, using both soft data and hard data as conditional data. The experimental results show that, compared with the unconditional reconstruction images and the reconstructed images using only hard data, the structure characteristics in reconstructed images using this method are similar to those obtained from real volume data.