Solar wind data that have been collected by earth-orbiting satellites are likely to be contaminated a significant part of the time by the proximity of the bow shock. Heating of the electron component in the shock-perturbed region is well established, whereas a temperature increase of the ion component has not been observed in the past. We presume that energy transfer to the solar wind protons in the magnetic field-bow shock connected region is effected by collisionless damping of hydromagnetic waves that probably are generated locally in this region by shock-reflected suprathermal protons. This process will influence only the temperature component T∥ parallel to the magnetic field direction and not the perpendicular component T⊥. The result, however, is not necessarily a pure increase in T∥, since such an increase may to some extent be reduced by a simultaneous expansion flow of the heated plasma. The physically most meaningful quantities for demonstrating the considered heating process are therefore T∥B2/n2 (where n denotes the proton number density and B the magnetic field intensity) and T⊥/B. In the absence of thermal energy transfer these parameters constitute adiabatic invariants of the anisotropic plasma. The heating process that is proposed as being operative in the shock-perturbed region should increase T∥B2/n2 and should leave T⊥/B unaffected. Actually, this is confirmed by our data. In this work we present a statistical study of the data measured aboard the Heos 2 satellite during the period February 1972 to August 1974. Employing T∥B2/n2 as an indicator, we investigate the geometry and extent of the upstream proton-heating zone and its relation to the bow shock. The heating process is found to be more effective nearer the shock than farther away. When we restrict the analysis to data that have been measured not more than 10 RE from the bow shock and calculate mean values, we find an increase of 47% in T∥B2/n2, an increase of 10% in T∥, and a simultaneous decrease of 13% in proton number density n for shock-perturbed conditions relative to unperturbed conditions. All the other parameters, especially T⊥/B, remain unaffected within statistical errors.