Fatigue life has a wide scatter and this fact makes fatigue design of structural members difficult. In order to make fatigue life distributions clear, it is required to prepare many number of specimens and to repeat the fatigue tests under the same loading condition. For such fatigue tests, it usually takes much time or cost. It is still more for fatigue tests such as random fatigue tests or corrosion fatigue tests in long life range. In this study, in order to reduce the time for experiment, a fatigue testing method using a multi-notched specimen which has many notches of the same shape and its reliability analysis method are developed. The fatigue test procedure by multi-notched specimen is as follows : First, loads are applied from the start of cycling up to certain cycles ( N 1). Then fatigue cracks are observed with microscope and the number of notches ( m 1) where fatigue cracks have initiated is counted. Again, loadings of certain cycles (Δ N ) are continued ( N 2= N 1+ Δ N ). Then the number of cracks ( m 2) newly initiated during Δ N cycles are counted. Repeating the process until the longest crack initiated in the specimen reaches the length that is defined before the test, a set of data ( N 1, m 1), ( N 2, m 2), …, ( N k, m k) is obtained as an event. The likelihood function of this event can be calculated employing multinomial distribution. Two parameter Weibull distributions are used as fatigue life distributions in this study. The Weibull parameters are estimated by the Bayesian reliability analysis or the likelihood ratio analysis. From the numerical calculations assuming a multi-notched specimen with sixty notches and the experiments using a multi-circular holed specimen, it is concluded that two or three multi-notched specimens are enough to estimate the fatigue life distributions and the time for test can be decreased below one-tenth comparing with the fatigue tests using a single notched specimen, with maintaining the reliabilities of the estimated fatigue life distributions.