We combine new HST/ACS observations and existing data to investigate the wavelength dependence of near-IR ( NIR ) extinction .
Previous studies suggest a power-law form for NIR extinction , with a “ universal ” value of the exponent , although some recent observations indicate that significant sight line-to-sight line variability may exist .
We show that a power-law model for the NIR extinction provides an excellent fit to most extinction curves , but that the value of the power , \beta , varies significantly from sight line-to-sight line .
Therefore , it seems that a “ universal NIR extinction law ” is not possible .
Instead , we find that as \beta decreases , R ( V ) \equiv A ( V ) / E ( B - V ) tends to increase , suggesting that NIR extinction curves which have been considered “ peculiar ” may , in fact , be typical for different R ( V ) values .

We show that the power law parameters can depend the wavelength interval used to derive them , with the \beta increasing as longer wavelengths are included .
This result implies that extrapolating power law fits to determine R ( V ) is unreliable .
To avoid this problem , we adopt a different functional form for NIR extinction .
This new form mimics a power law whose exponent increases with wavelength , has only 2 free parameters , can fit all of our curves over a longer wavelength baseline and to higher precision , and produces R ( V ) values which are consistent with independent estimates and commonly used methods for estimating R ( V ) .
Furthermore , unlike the power law model , it gives R ( V ) s that are independent of the wavelength interval used to derive them .
It also suggests that the relation R ( V ) = -1.36 \frac { E ( K - V ) } { E ( B - V ) } -0.79 can estimate R ( V ) to \pm 0.12 .

Finally , we use model extinction curves to show that our extinction curves are in accord with theoretical expectations , and demonstrate how large samples of observational quantities can provide useful constraints on the grain properties .