Motivated by several recent data, we test the QCD spectral sum rules (QSSR) predictions based on different proposals (\bar qq, \bar q\bar q qq, and gluonium) for the nature of scalar mesons. In the I=1 and 1/2 channels, the unusual (wrong) splitting between the a_0(980) and \kappa(900) and the a_0(980) width can be understood from QSSR within a \bar qq assignement. However, none of the \bar qq and \bar q\bar q qq results can explain the large \kappa width, which may suggest that it can result from a strong interference with non-resonant backgrounds. In the I=0 channel, QSSR and some low-energy theorems (LET) require the existence of a low mass gluonium \sigma_B(1 GeV) coupled strongly to Goldstone boson pairs which plays in the U(1)_V channel, a similar role than the \eta' for the value of the U(1)_A topological charge. The observed \sigma(600) and f_0(980) mesons result from a maximal mixing between the gluonium \sigma_B and \bar qq(1 GeV) mesons, a mixing scheme which passes several experimental tests. OZI violating J/\psi--> \phi\pi^+\pi^-, D_s--> 3\pi decays and J/\psi--> \gamma S glueball filter processes may indicate that most of the I=0 mesons above 1 GeV have important gluonium in their wave functions. We expect that the f_0(1500), f_0(1710) and f_0(1790) have significant gluonium component in their wave functions, while the f_0(1370) is mostly \bar qq. Tests of these results can be provided by the measurements of the pure gluonium \eta'\eta and 4\pi specific U(1)_A decay channels.