The application of texture analysis (TA) in magnetic resonance imaging (MRI) requires the availability of texture phantoms for use in the standardization of in vivo measurements. The aims of our study were (a) to develop a new type of phantoms suitable for MRI and TA and test their long-term stability; (b) to optimize the choice of texture parameters describing the phantoms; (c) to compare different MR imagers according to texture parameters in a multi-center study. A long-term study performed at 4.7 T proved that the developed phantom based on polystyrene spheres and an agar gel solution is stable at least 12 months. This phantom, with nodular patterns, was found useful for the modeling of structural differences. The comparison of TA parameters at 4.7 and 7 T proved that the same parameters can be used for the separation of structures. The proposed algorithm of the selection of TA parameters shows that there exists a part of texture parameters which can be measured with high reproducibility (1-3%); on the other hand, their absolute values can differ by more than 30% if the textures differ. Results obtained from the multi-center study of whole body MR imagers show the wide variation in the misclassification rates at the different sites and point out the importance of the set up of MR sequences.