The POVM is a set of operators ${E_m}$, that are positive and $\sum_m{E_m} = \mathbb{1}$. The probability for a measurement outcome $m$ is then given by $p(m) = \braket{\phi|E_m|\phi}$.