The parameter space must be finite.

The support of the density must depend on $\theta$.

The log-likelihood must be twice differentiable with respect to $\theta$.

The data must be normally distributed.