Records of diversion and return flows for water years 1961?2004 along a reach of the Klamath River between Link River and Keno Dams in south-central Oregon were evaluated to determine the cause of a water-balance inconsistency in the hydrologic data. The data indicated that the reach was losing flow in the 1960s and 1970s and gaining flow in the 1980s and 1990s. The absolute mean annual net water-balance difference in flows between the first and second half of the 44-year period (1961-2004) was approximately 103,000 acre-feet per year (acre-ft/yr). The quality of the diversion and return-flow records used in the water balance was evaluated using U.S. Geological Survey (USGS) criteria for accuracy. With the exception of the USGS Klamath River at Keno record, which was rated as 'good' or 'excellent,' the eight other flow records, all from non-USGS flow-measurement sites, were rated as 'poor' by USGS standards due to insufficient data-collection documentation and a lack of direct discharge measurements to verify the rating curves. The record for the Link River site, the most upstream in the study area, included both river and westside power canal flows. Because of rating curve biases, the river flows might have been overestimated by 25,000 acre-ft/yr on average from water years 1961 to 1982 and underestimated by 7,000 acre-ft/yr on average from water years 1983 to 2004. For water years 1984-2004, westside power canal flows might have been underestimated by 11,000 acre-ft/yr. Some diversion and return flows (for mostly agricultural, industrial, and urban use) along the Klamath River study reach, not measured continuously and not included in the water-balance equation, also were evaluated. However, the sum of these diversion and return flows was insufficient to explain the water-balance inconsistency. The possibility that ground-water levels in lands adjacent to the river rose during water years 1961-2004 and caused an increase in ground-water discharge to the river also was evaluated. However, water-level data from local wells did not have a rising trend during the period. The most likely cause of the water-balance inconsistency was flow measurement error in the eight non-USGS flow records. Part of the water-balance inconsistency can be explained by a 43,000 acre-foot error in the river and canal flow portions of the Link River flow record. A remaining 60,000 acre-foot error might have been distributed among the seven other flow records, or much of the remaining 60,000 acre-foot error might have been in the Link River flow record because flows in that record had a greater magnitude than flows in the seven other records. As an additional analysis of the water-balance issue, flow records used in the water balance were evaluated for trends and compared to known changes in water management in the Bureau of Reclamation Klamath Project and Lower Klamath and Tule Lake National Wildlife Refuges over the 44-year period. Many of the water-management changes were implemented in the early 1980s. For three diversion flow records, 1983-2004 mean annual flows were 16,000, 8,000, and 21,000 acre-ft/yr greater than their 1961-82 mean annual flows. Return flows to the Klamath River at two flow-measurement sites decreased by 31,000 and 27,000 acre-ft/yr for 1983-2004 compared with the 1961-82 period.