Many experiments indicate that isospin is not a good quantum number for the states with Jπ=2+ at 16.6 MeV and 16.9 MeV in Be8. The available experimental information is examined and compared with a model of single particle states which are maximally mixed in isospin, as proposed by MARION. An analysis of all data pertinent to the states at 17.6 MeV and 18.15 MeV with Jπ=1+ shows that these states are rather pure (90%) isospin states. The state at 17.6 MeV has T=1, the one at 18.15 MeV hat T=0. Both the 2+ and the 1+ levels form doublets with similar wave functions in each case. A similar pair is most likely formed by the 3* states at 19.05 MeV and 19.22 MeV, as proposed by BARKER. Reduced widths indicate possible isospin mixing of as much as 30% for these states. Some experimental evidence is presented for another pair of states with Jπ=1- at 20.35 MeV and 22 MeV, and with Jπ=2- at 18.9 MeV and 23.6 MeV. This hypothesis provides analog states in Be8 for every state observed in Li8 in the energy region which is considered. If each pair (except the 1+ and 2+ pairs) is allowed to mix isospin by at most 30%, all cases of large excess of either neutron or proton reduced width can be explained. The large isospin mixture in the 2* pair can be explained on the basis of an initial energy degeneracy of these states (removed by the COULOMB force). The intermediate coupling shell model does account for such an accident. It is argued that, although a simple cluster picture of these states is highly successful in a qualitative way, the intermediate coupling model is able and probably needed to achieve a quantitative description of Be8 between 16 and 20 MeV.