New Monte Carlo estimators of the effective delayed neutron fraction eff are presented in this paper. By looking at the physical interpretation of the adjoint function, one can incorporate its effect on the delayed neutron fraction without explicitly calculating the adjoint function itself. We have implemented these estimators into MCNP. In a standard keff calculation, the code now reports a eff value. The method does not slow down the code by more than 0.5%. We propose an extensive experimental benchmark set for eff, which we use to test our method and two known approximate methods. Our method reproduces all experimental values.