The effect of full-scale model tests, or comparable ones, to the reliability assessment of a structure was discussed assuming that (1) the strength of the underlying structure follows a three-parameter Weibull distribution, (2) its characteristic value is unknown, (3) the test load is limited to a certain value because of the capacity of testing facility or the testing scheme, and (4) a Bayesian analysis is applied to the test results. General trends of the Bayesian reliability obtained by such a procedure were discussed focusing on the effect of the maximum testing load, number of samples avairable and the prior estimation of the structural reliability. Main remarks are : (1) Tests have, regardless of the sample size, no value if the test load is too low compared with the expected service load. (2) It may be expected that the higher the test load is, the more effective it is. When sample size is very small, this may not be true. (3) To draw effective informations from the test, including the sellection of an ample combination of the sample size and the test load, a reasonable, neither conservative nor unconcervative, prior distribution of the unknown parameter, characteristic value in the present study, is to be chosen. (4) On chosing a prior distribution, the uncertainty of the other parameters, if any, should be taken into consideration as well as the required reliability. (5) A rather small shape factor is recommended if there is any uncertainty on it, since misestimation of a shape factor possibly reduces the effect of a test. (6) Any number of successive successes in supplementary tests, which must be done after getting a fail at the first test to demonstrate enough reliability can be expected, have no effect at all if test load is too low. (7) A supplementary test with a higher load than the previous test may produce sufficient results even if the successive successes are few after the fail in the first test.