The loss of both habitat quality and quantity for anadromous fish in the Columbia River basin has been identified as a major factor in the decline of many species and has been linked to a variety of land management activities. In this study, we compared stream reaches in watersheds representing both managed and reference conditions to determine whether we could detect differences in physical habitat variables. We divided stream habitat measures into three components: Stream banks, instream habitat (pools and pool depth), and stream substrate. We randomly sampled perennial streams within 261 sixth hydrologic unit code (HUC) stream reaches on federal lands in Idaho, Montana, Oregon, and Washington. The sample population represented stream reaches in 62 reference watersheds and 199 managed watersheds. An unbalanced, incomplete-block-design analysis of covariance (ANCOVA) was performed on each of the habitat variables using geology type as the block effect and bank-full width, stream gradient, and average precipitation as the covariates. Watersheds containing reference stream reaches had a slightly higher percentage of federal lands, were smaller, tended to occur at higher mean elevations, and received more annual precipitation than did the managed watersheds. We observed differences in most measures of stream habitat between reference and managed watersheds, generally in the direction we expected. Stream banks were more stable and more undercut in reference stream reaches. Pools in reference stream reaches were deeper than pools in managed stream reaches and had less fine sediment in pool tails. Analysis of covariance was an effective way to compare data across a large, relatively heterogeneous landscape where sample site stratification may be impractical or sample sizes are limited. We believe that the comparison of reference conditions to conditions across managed landscapes provides a credible way to report on the condition of these systems in lieu of trend information at individual sites.