A precise prediction of expected neutrino fluxes is required for a long-baseline accelerator neutrino experiment. The flux is used to measure neutrino cross sections at the near detector, while at the far detector it provides an estimate of the expected signal for the study of neutrino oscillations. In the talk several approaches to constrain the ν flux are presented. The first is the traditional one when an interaction chain for the neutrino parent hadrons is stored to be weighted later with real measurements. In this approach differential hadron cross sections are used which, in turn, are measured in ancillary hadron production experiments. The approach is certainly model dependent because it requires an extrapolation to different incident nucleon momenta assuming xF scaling as well as extrapolation between materials having different atomic numbers. In the second approach one uses a hadron production yields off a real target exploited in the neutrino beamline. Yields of neutrino parent hadrons are parametrized at the surface of the target, thus one avoids to trace the particle interaction history inside the target. As in the case of the first approach, a dedicated ancillary experiment is mandatory. Recent results from the hadron production experiments -NA61/SHINE at CERN (measurements for T2K) and MIPP at Fermilab (measurements for NuMI) -are reviewed.