National Ignition Facility (NIF) specifications have stringent dimensional accuracy requirements on target components. For example, the laser-hole diameter on an ablator capsule must be characterized to ±0.5 m to ensure proper fill tube insertion and to minimize the glue joint mass to <2.5 ng. A charge-coupled-device-based X-ray radiography and tomography instrument (commercially obtained from Xradia, Inc.) is used in target metrology where sample opacity precludes the use of optical techniques; however, the built-in caliper for dimensional measurement cannot provide the required accuracy. The instrument has three main error sources: (a) point projection magnification, (b) imaging lens distortion, and (c) phase contrast shift. The sample feature size dictates the calibration strategy. For large features such as the shell diameter, (a) and (b) dominate the error budget. The built-in caliper is accurate to ~2 to 3%, corresponding to a ±50-m error for a 2000-m NIF capsule. In this work, we developed an X-ray transmission dimension standard and developed (by measuring the standard) a software algorithm to "un-distort" the acquired images without resorting to the standard each time. The latter approach reduces the processing time by 50% and still offers a tenfold accuracy improvement and makes the Xradia instrument useful in screening components. For small features such as laser-drilled holes, (c) is dominant. It shifts the apparent wall boundary to cause a typical ~2-m error for the 5- to 10-m hole diameter. We developed an empirical correction technique with 0.5-m accuracy, in which the dimensions measured by radiography were benchmarked against those by a focused ion beam and scanning electron microscope after sample cleavage. The improved accuracy allows the glue mass to be estimated to 1 ng as required by the NIF specifications.