Mergers of double neutron stars are considered the most likely progenitors for short gamma-ray bursts. Indeed, such a merger can produce a black hole with a transient accreting torus of nuclear matter, and the conversion of a fraction of the torus mass-energy to radiation can power a gamma-ray burst. Using available binary pulsar observations supported by our extensive evolutionary calculations of double neutron star formation, we demonstrate that the fraction of mergers that can form a black hole-torus system depends very sensitively on the (largely unknown) maximum neutron star mass. We show that the available observations and models put a very stringent constraint on this maximum mass under the assumption that black hole formation is required to produce a short gamma-ray burst in a double neutron star merger. Specifically, we find that the maximum neutron star mass must be within . Moreover, a single unambiguous measurement of a neutron star mass above 2-2.5 M , would exclude a black hole-torus central engine model of short gamma-ray bursts in double neutron 2.5 M , star mergers. Such an observation would also indicate that if in fact short gamma-ray bursts are connected to neutron star mergers, the gamma-ray burst engine is best explained by the lesser known model invoking a highly magnetized massive neutron star. Subject headings: binaries: close -black hole physics -gravitational waves -stars: evolutionstars: neutron Online material: color figures Gamma-ray bursts (GRBs) have been separated into two classes: long-soft bursts and short bursts (Nakar 2007;Gehrels et al. 2007). The origin of long-soft bursts has been connected to the death of low-metallicity massive stars (Piran 2005;Gehrels et al. 2007). However, while observations support a binary merger origin for short bursts (Nakar 2007;Gehrels et al. 2007), the exact nature of the progenitor remains uncertain: they could be either double neutron stars (NS-NS) or black hole-neutron star (BH-NS) binaries. The number of BH-NS binaries that both merge and produce GRBs is hard to estimate since (i) no such system has yet been observed, (ii) formation models are rather uncertain and predict very small BH-NS merger rates (likely too small to explain most of the short bursts), and (iii) theory suggests that the fraction of BH-NS mergers producing bursts depends sensitively on the black hole spin and spin-orbit orientation (Belczynski et al. 2008b), but black hole birth spins are not well constrained observationally or theoretically. On the other hand, NS-NS binaries are observed only in the Milky Way, but their properties and numbers are also in agreement with theoretical models, and their merger rate is sufficient to explain the present-day short-burst population (Nakar 2007;Belczynski et al. 2007).We have performed an extensive theoretical study of highmass binary stars (potential progenitors of NS-NS systems) using StarTrack, a population synthesis code incorporating the most up-to-date and detailed input physics for massive stars (Belczynski et al. ...