In this note we introduce the concept of local average dimension of a
measure mu at x is an element of R(n) as the unique exponent where the
lower average density of mu at x jumps from zero to infinity. Taking
the essential infimum or supremum over x we obtain the lower and upper
average dimensions of mu, respectively. The average dimension of an a
nalytic set E is defined as the supremum over the upper average dimens
ions of all measures supported by E. These average dimensions lie betw
een the corresponding Hausdorff and packing dimensions and the inequal
ities can be strict. We prove that the local Hausdorff dimensions and
the local average dimensions of mu at almost all x are invariant under
orthogonal projections onto almost all m-dimensional linear subspaces
of higher dimension. The corresponding global results for mu and E (w
hich are known for Hausdorff dimension) follow immediately.