A necklace can be considered as a cyclic list of $n$ red and $n$ blue beads in an arbitrary order, and the goal is to fold it into two and find a large cross-free matching of pairs of beads of different colors. We give a counterexample for a conjecture about the necklace folding problem, also known as the separated matching problem. The conjecture (given independently by three sets of authors) states that $μ=\frac{2}{3}$, where $μ$ is the ratio of the `covered' beads to the total number of beads. We refute this conjecture by giving a construction which proves that $μ\le 2 \nolinebreak - \nolinebreak \sqrt 2 < 0.5858$. Our construction also applies to the homogeneous model: when we are matching beads of the same color. Moreover, we also consider the problem where the two color classes not necessarily have the same size.