It is suggested that inference under the proportional hazard model can be carried out by programs for exact inference under the logistic regression model. Furthermore a different type of exact inference is developed under Type II censoring. Performance of logistic exact and exact inference is compared to large sample Wald, score and likelihood inference by coverage and power calculations. The logistic exact confidence intervals have coverage well above the nominal level in most computation. The exact inference under Type II censoring turn out to be less conservative but numerically very complex. Large sample methods works well with remarkably small data but score and likelihood ratio methods are preferable to inference by Wald statistics.