Two new methods are used to obtain data on neutron yield ${\ensuremath{\nu}}_{f}$ as a function of the mass number $M$ of the emitting fission fragment; the results suggest changes in fission theory. The methods involve the combination of data on fission mass yields obtained by radiochemical means with time-of-flight mass data. The more detailed method yields essentially the complete function ${\ensuremath{\nu}}_{f}(M)$. Both methods are applied to neutron fission of ${\mathrm{U}}^{233}$, ${\mathrm{U}}^{235}$, and ${\mathrm{Pu}}^{239}$, and to spontaneous fission of ${\mathrm{Cf}}^{252}$. The neutron yields obtained, which show strong variation with mass, are compared with other available data; no previous results exist for ${\mathrm{Pu}}^{239}$. The four sets of data on ${\ensuremath{\nu}}_{f}(M)$ are so similar that it is suggested that the neutron yield is primarily a function of fragment mass and not of mass ratio. In each case minimum neutron yield is found for $N\ensuremath{\cong}50$ and $Z=\ensuremath{\cong}50$ (masses 82 and 128). It is suggested that these magic and near-magic fragments have low excitations, and consequently emit almost no neutrons, because of greater rigidity against distortion from near-spherical shapes. The same idea leads to a prediction of high fission barrier and low fission yield at these masses, which compares well with data for single-peaked, triple-peaked, and ordinary double-peaked-mass-yield fission. In the appendices a number of necessary relations are developed, concerning neutron emission energies. The width of the neutron emission spectrum is deduced from laboratory spectra; new relations between emission spectra and angular correlation of neutrons and fragments are developed and applied; and it is shown that the extra width of fragment mass yields obtained from fragment energies (semiconductor or ionization-chamber data) is due to the correlation of ${\ensuremath{\nu}}_{f}$ and $M$.