Beach erosion is ubiquitous along the U. S. East Coast— ∼80–90% of the beaches are eroding. Federal and state agencies thus expend a great deal of effort to determine erosion rates for establishment of construction setback lines. Historical shoreline positions are used to calculate rates of change of beach width, but the temporal variability of shoreline position creates difficulties. In most places that are highly developed or likely to become so, the annual rate of beach erosion (order ∼1 meter) is small compared to the accuracy of shoreline position measurement (order ∼10 meters) or the seasonal-to-interannual and longer fluctuations of beach width (order tens of meters). This unfavorable signal-to-noise ratio makes determining the underlying long-term rate of erosion problematic from even half-century- long shoreline position records unless great care is taken. Making useful shoreline position predictions and their associated errors requires an understanding of the sources of temporal variability of shoreline position. We have used real shoreline position data in endpoint rate (difference of two shoreline positions divided by time) and linear regression analyses to demonstrate the essential features of the problem for several U.S. East Coast shorelines. The scatter in computed end point rates is so large at time scales <60–80 years that an arbitrary end point rate trend is as likely to be erosional as accretional, demonstrating that the end point rate method should not be used. Linear regression usually leads to much smaller errors in shoreline change rate, but significant errors in predicted position and especially the uncertainty of the prediction will result if storminfluenced shoreline positions are included in the computation.