First and second normal stress differences of a 50 vol.% magnetorheological fluid (MRF) are investigated by using a commercial plate-plate magneto-rheometer (Anton Paar GmbH) with plate-plate and cone-plate geometry. The manufacturer modified the instrument to achieve higher normal force (60 N) and torque (295 mNm) capacity. An additional modification by us allows an online determination of the true magnetic flux density B in the MRF by means of a Hall probe. FEM Maxwell 2D simulations quantitatively verify the Hall probe results and give detailed insight into the radial flux density profile within the MRF sample. Without shear, the static normal force F-N for plate-plate increases as a power law: F-N proportional to B-2.4. A similar magnitude is found for cone-plate geometry, in contrast to the expectation. For steady shear at 10s(-1) the plate-plate normal force built-up limits the experiments at high flux densities rather than the torque generated. The normal forces increase linearly with the shear stress at high flux density. The first normal stress difference N, is positive and about five times larger than the shear stress. The second normal stress difference N-2 is also positive. The experimentally derived N-2/N-1 ratio of 1/4 distinctly deviates from theoretical predictions (N-2/N-1 = -1) for a semi-dilute MRF. Improvements of the radial flux density profiles are required to verify the normal stress difference ratio and to support the conjecture that the positive but small N-2 is a consequence of the densely packed MRF, which does not allow to create extended chain-like structures. As shown in the appendix, the experimentally determined N-2/N-1 ratio is favorable to stabilize concentricity in concentric cylinder arrangements, relevant for the MRF application in clutches. (c) 2007 Elsevier B.V. All rights reserved.