Are you talking about a vector or should this be applied to columns of a matrix?
For a vector:
x = rand(100, 1);
x(17:100) = 0;
idx = find(x == 0, 1);
if ~isempty(idx)
result = sum(x(1:idx - 1)) / (idx - 1);
else % No zero found:
result = sum(x) / length(x); % Or mean() of course
end
For a matrix you could work with Idx = cumsum(X == 0, 1) == 0. But creating the intermediate index matrix will most likely need more time than the vectorization saves. Therefore I'd stay at the above method for vectors and put it into a loop over the columns.
