Using the properties of the quasi-stellar object (QSO) catalogue of Shen et al., from the Seventh Data Release of the Sloan Digital Sky Survey, we investigate the Baldwin effect, its slope evolution, and the underlying drive for a large sample of 35 019 QSOs with reliable spectral analysis. We find that the Baldwin effect exists in this large sample of QSOs, and that it is almost the same in 11 different redshift bins, up to z ∼ 5. The slope is −0.238 using the bivariate correlated errors and intrinsic scatter (BCES) algorithm – the equivalent width (EW) of C iv depends on the continuum – and −0.787 using the BCES bisector. For 11 redshift bins, there is an increase in the slope of the Baldwin effect from z ∼ 1.5 to z ∼ 2.0. From z ∼ 2.0 to z ∼ 5.0, the change in the slope is not clear, considering the uncertainties or larger redshift bins. There is a strong correlation between the rest-frame C iv EW and the C iv-based MBH, while the relation between the C iv EW and the Mg ii-based MBH is very weak. With the correction of the C iv-based MBH from the C iv blueshift relative to Mg ii, we suggest that this strong correlation is a result of the bias of the C iv-based MBH, with respect to that from the Mg ii line. Considering the Mg ii-based MBH, a medium strong correlation is found between the C iv EW and the Eddington ratio, which implies that the Eddington ratio seems to be a better underlying physical parameter than the central black hole mass.