The NATRE fine- and microstructure data set is revisited to test salt-finger amplitude theories. Dependences of the mixing efficiency ?, microscale buoyancy Reynolds number Re and thermal Cox number CxT on 5-m density ratio R? and gradient Richardson number Ri are examined. The observed mixing efficiency is too high to be explained by linear fastest-growing fingers but can be reproduced by wavenumbers 0.5-0.9 times lower than the fastest-growing wavenumber. Constraining these fingers with a hybrid wave/finger Froude number or a finger Reynolds number cannot reproduce the observed trends with R? or Ri, respectively. This suggests that background shear has no influence on finger amplitudes. Constraining average amplitudes of these lower-wavenumber fingers with finger Richardson number Rif ~ 0.2 reproduces the observed dependence of Re and CxT on density ratio R? and Ri at all but the lowest observed density ratio (R? = 1.3). Separately relaxing the assumptions of viscous control, dominance of a single mode and tall narrow fingers does not explain the difference between theory and data at low R? for a critical Rif ~ 0.2.