Context: We recently reported on the detection of a possible planetary mass companion to β Pictoris at a projected separation of 8 AU from the star, using data taken in November 2003 with NaCo, the adaptive-optics system installed on the Very Large Telescope UT4. Even though no second epoch detection was available, there are strong arguments to favor a gravitationally bound companion rather than a background object. If confirmed and located at a physical separation of 8 AU, this companion would be the closest planet ever imaged, and above all, could have formed via core-accretion. Its apparent magnitude would indicate a typical temperature of ~1500 K and a mass of ~8 MJup. Interestingly, a planet with such characteristics would explain the main morphological and dynamical peculiarities of the β Pic system. Aims: Our goal was to re-observe β Pic five years later to again detect the companion or, in the case of a non-detection, constrain its orbit. Methods: Deep adaptive-optics L'-band direct images of β Pic as well as K_s-band Four Quadrant Phase Mask coronagraph images with were recorded with NaCo. Results: No point-like signal with the brightness of the companion candidate (apparent magnitudes L' = 11.2 or Ks ≃ 12.5) was detected at projected distances down to ≃6.5 AU in the present data from the star (by comparison, the same limit was reached at ≃5.5 AU in the better quality November 2003 data). As expected, the non detection does not allow us to rule out a background companion from an observational point of view. We show that the non detection is consistent with orbital motion. Using these data and previous K_s-band data obtained in 2004, we place strong constraints on the possible orbits of the companion. Based on observations collected at the European Southern Observatory, Chile, ESO; runs 282.C5037(A), 282.C5037(B) and 282.C5037(D).