We empirically MD test the relation between the SFR(LIR) derived from the infrared luminosity, LIR, and the SFR(Hα) derived from the Hα emission line luminosity using simple conversion relations. We use a sample of 474 galaxies at z = 0.06–0.46 with both Hα detection [from 20k redshift Cosmological Evolution (zCOSMOS) survey] and new far-IR Herschel data (100 and 160 μm). We derive SFR(Hα) from the Hα extinction corrected emission line luminosity. We find a very clear trend between E(B − V) and LIR that allows us to estimate extinction values for each galaxy even if the Hβ emission line measurement is not reliable. We calculate the LIR by integrating from 8 up to 1000 μm the spectral energy distribution (SED) that is best fitting our data. We compare the SFR(Hα) with the SFR(LIR). We find a very good agreement between the two star formation rate (SFR) estimates, with a slope of m = 1.01 ± 0.03 in the log SFR(LIR) versus log SFR(Hα) diagram, a normalization constant of a = −0.08 ± 0.03 and a dispersion of σ = 0.28 dex. We study the effect of some intrinsic properties of the galaxies in the SFR(LIR)–SFR(Hα) relation, such as the redshift, the mass, the specific star formation rate (SSFR) or the metallicity. The metallicity is the parameter that affects most the SFR comparison. The mean ratio of the two SFR estimators log[SFR(LIR)/SFR(Hα)] varies by ∼0.6 dex from metal-poor to metal-rich galaxies [8.1 < log (O/H) + 12 < 9.2]. This effect is consistent with the prediction of a theoretical model for the dust evolution in spiral galaxies. Considering different morphological types, we find a very good agreement between the two SFR indicators for the Sa, Sb and Sc morphologically classified galaxies, both in slope and in normalization. For the Sd, irregular sample (Sd/Irr), the formal best-fitting slope becomes much steeper (m = 1.62 ± 0.43), but it is still consistent with 1 at the 1.5σ level, because of the reduced statistics of this sub-sample.