Under current social security provisions and leading reform proposals, benefits of retirees eventually will be based on an average of the 35 highest annual taxable earnings figures. At present, the averaging period is considerably shorter. However, the number of dropout years in the averaging process will be unchanged. For different expected lifetime patterns and different stochastic specifications of earnings, the paper estimates the effect of lengthening the averaging period on the expected value of lifetime average earnings. The effects of lengthening the period are significant and vary noticeably with both expected wages and stochastic specification. Under current social security provisions, benefits of retirees eventually will be based on an average of the 35 highest annual taxable earnings figures. Proposals to index earnings, before selecting the years to be used in determining benefits, generally have preserved this long averaging period. However, at present, earnings before 1951 generally are not used in calculating benefits. Adjusting for the limited availability of earnings records, current law provides for a shorter averaging period for normal retirements (at age 65) until 1994. In 1978 when 65 year old male and female retirees will have the same averaging period, its length will be 19 years. Thus, for an individual retiree at age 65, benefits will be based on the best 19 out of 27 years of earnings possibly included in the calculation. The averaging period is scheduled to increase one year every year until it reaches 35. For a retiree who is just 65 and started work at 21, this represents selection of the best 35 years out of 43 full years of earnings.1 This lengthening averaging period will imply different benefits for retirees, born at different times, despite the fact that they may have had the same (indexed) earnings history. (Some reform proposals make benefits vary with date of birth or retirement in other ways too.) The lengthening Yves Balcer is an Assistant Professor in the Faculty of Management Sciences at the University of Ottawa and a Ph.D. candidate in economics at MIT. Peter Diamond, Ph.D., is a Professor of Economics at MIT. Financial support by NSF is gratefully acknowledged. 1 Those starting full time work earlier or retiring later have additional years of earnings available for averaging. The effects of bad earnings years on the propensity to retire early and the possibility that years of partial earnings will show higher earnings than years of full earnings have been ignored.