The v=1 and 2 J=1-0 (43GHz), and v=1 J=2-1 (86GHz) SiO masers are intense in AGB stars and have been mapped using VLBI showing ring-like distributions. Those of the v=1, 2 J=1-0 masers are similar, but the spots are rarely coincident, while the v=1 J=2-1 maser arises from a well separated region farther out. These relative locations can be explained by models including the overlap of two IR lines of SiO and H_2_O. The v=3 J=1-0 line is not directly affected by any line overlap and its spot structure and position, relative to the other lines, will be a good test to the standard pumping models The aims is to gain insight into the properties of the different SiO masers and the general theoretical considerations that can help to understand them. We present single-dish and VLBI simultaneous observations of the v=1, 2, 3 J=1-0 maser transitions of ^28^SiO in several AGB stars. The results are compared with the predictions of radiative models of SiO masers both including and not including the effect of IR line overlap. The spatial distribution of the SiO maser emission in the v=3 J=1-0 transition from AGB stars is systematically composed of a series of spots occupying a ring-like structure (as often found in SiO masers). The overall ring structure is extremely similar to that found in the other 43 GHz transitions, and very different from the structure of the v=1 J=2-1 maser. The positions of the individual spots of the different 43 GHz lines are however very rarely coincident, being in general separated by about 0.3AU (between 1 and 5mas). These results are very difficult to reconcile with standard pumping models, which predict the masers of rotational transitions within a given vibrational state to require very similar excitation conditions (since the levels are placed practically at the same energy from the ground), while the transitions of different vibrational states (which are separated by an energy of 1800K) should appear in different positions. However, models including line overlap tend to predict v=1, 2, 3 J=1-0 population inversion to occur under very similar conditions, while the requirements for v=1 J=2-1 appears clearly different, and are compatible with the observational results.