The latest cosmological data seem to indicate a significant deviation from scale invariance of the primordial power spectrum when parameterized either by a power law or by a spectral index with non-zero running. This deviation, by itself, serves as a powerful tool to discriminate among theories for the origin of cosmological structures such as inflationary models. Here, we use a minimally-parametric smoothing spline technique to reconstruct the shape of the primordial power spectrum. This technique is well-suited to search for smooth features in the primordial power spectrum such as deviations from scale invariance or a running spectral index, although it would recover sharp features of high statistical significance. We use the WMAP 3 year results in combination with data from a suite of higher resolution CMB experiments (including the latest ACBAR 2008 release), as well as large-scale structure data from SDSS and 2dFGRS. We employ cross-validation to assess, using the data themselves, the optimal amount of smoothness in the primordial power spectrum consistent with the data. This minimally-parametric reconstruction supports the evidence for a power law primordial power spectrum with a red tilt, but not for deviations from a power law power spectrum. Smooth variations in the primordial power spectrum are not significantly degenerate with the other cosmological parameters.