I found this on wikipedia:
Effectiveness of ECPs
The current FDA-approved U.S. product labeling states that levonorgestrel treatment can prevent 89% of expected pregnancies, and that EC (including the Yuzpe method) reduces the risk of pregnancy by at least 75%.[14]
The effectiveness of emergency contraception is expressed as a percentage reduction in pregnancy rate for a single use of EC. An article in American Family Physician explains a 75% effectiveness rate thus:
... these numbers do not translate into a pregnancy rate of 25 percent. Rather, they mean that if 1,000 women have unprotected intercourse in the middle two weeks of their menstrual cycles, approximately 80 will become pregnant. Use of emergency contraceptive pills would reduce this number by 75 percent, to 20 women.[15]
The effectiveness of emergency contraception is highest when taken within 12 hours of intercourse and declines over time.[16][17] The limit of 72 hours is based on a study by the World Health Organization (WHO).[7] A subsequent WHO study has suggested that reasonable effectiveness continues for up to 120 hours (5 days) after intercourse.[18] However, many doctors (particularly in the U.K.) advise use of an IUD rather than ECP's for emergency contraception between 72 and 120 hours.
Uncertainties in calculation
The original method of calculating the failure rate of EC was to divide the number of observed pregnancies by the number of women treated. Typically, fewer than 1 in 100 women in these studies became pregnant, resulting in published effectiveness rates of 99%. However, a 1980 paper first used an alternate method: observed pregnancies were divided by the estimated number of women who would have become pregnant without the treatment.[19] This method was seen as more useful and quickly gained popularity; it is used in all studies of ECPs today.
Studies of emergency contraception have found widely varying results: 55%-94% effectiveness rates have been reported for the Yuzpe method.[15] Levonorgestrel is more effective than Yuzpe by 36%[15]-49%.[20]
Placebo-controlled trials that could give a precise measure of effectiveness for EC would be unethical, so the effectiveness percentage is estimated. This is currently done using variants of the calendar method.[21] Women with irregular cycles for any reason (including recent hormone use such as oral contraceptives and breastfeeding) must be excluded from such calculations. Even for women included in the calculation, the limitations of calendar methods of fertility determination have long been recognized. Recently, hormonal assay has been suggested as a more accurate method of estimating fertility for EC studies.[22]