We present evidence that the mid infrared (MIR) is a good tracer of the total infrared luminosity, L(IR), and star formation rate (SFR), of galaxies up to z 1.3. We use deep MIR images from the Infrared Space Observatory (ISO) and the Spitzer Space Telescope in the Northern field of the Great Observatories Origins Deep Survey (GOODS-N) together with VLA radio data to compute three independant estimates of L(IR). The L(IR,MIR) derived from the observed 15 and/or 24 um flux densities using a library of template SEDs, and L(IR,radio), derived from the radio (1.4 and/or 8.5 GHz) using the radio-far infrared correlation, agree with a 1-sigma dispersion of 40 %. We use the k-correction as a tool to probe different parts of the MIR spectral energy distribution (SED) of galaxies as a function of their redshift and find that on average distant galaxies present MIR SEDs very similar to local ones. However, in the redshift range z= 0.4-1.2, L(IR,24um) is in better agreement with L(IR,radio) than L(IR,15 um) by 20 %, suggesting that the warm dust continuum is a better tracer of the SFR than the broad emission features due to polycyclic aromatic hydrocarbons (PAHs). We find marginal evidence for an evolution with redshift of the MIR SEDs: two thirds of the distant galaxies exhibit rest-frame MIR colors (L(12 um)/L(7 um) and L(10 um)/L(15 um) luminosity ratios) below the median value measured for local galaxies. Possible explanations are examined but these results are not sufficient to constrain the physics of the emitting regions. We compare three commonly used SED libraries which reproduce the color-luminosity correlations of local galaxies with our data and discuss possible refinements to the relative intensities of PAHs, warm dust continuum and silicate absorption.

15 pages, 9 figures, accepted for publication in A&A