I test the completeness of USNO-B detections of high proper-motion ( \mu > 180 { mas } { yr } ^ { -1 } ) stars and the accuracy of its measurements by comparing them to the revised NLTT ( rNLTT ) catalog of Salim & Gould .
For 14.5 < V < 18.5 , only 6 % of such stars are missing from USNO-B while another 4 % have large errors , mostly too large to be useful .
Including both classes , incompleteness is 10 % .
These fractions rise toward both brighter and fainter magnitudes .
Incompleteness rises with proper motion to \sim 30 \% at \mu = 1 ^ { \prime \prime } yr ^ { -1 } .
It also rises to \sim 35 \% at the Galactic plane , although this is only determined for relatively bright stars V \lesssim 14 .
For binaries , incompleteness rises from 10 % at separations of 30 ^ { \prime \prime } to 47 % at 10 ^ { \prime \prime } .
The proper-motion errors reported internally by USNO-B are generally correct .
However , there is floor of \sigma _ { \mu } \sim 4 { mas } { yr } ^ { -1 } below which the reported errors should not be taken at face value .
The small number of stars with relatively large reported errors ( \sigma _ { \mu } \gtrsim 20 { mas } { yr } ^ { -1 } ) may actually have still larger errors than tabulated .