Situations in which the name of a person is perceived as familiar but does not trigger recall of pertinent semantic knowledge are common in daily life. In current connectionist models of person recognition, such "familiar-only" experiences reflect supra-threshold activation at person-identity nodes but subthreshold activation at nodes representing semantic knowledge. As knowledge is posited to be either present or absent according to a threshold, these models predict that no semantic knowledge should be observed in association with familiar-only experiences. In 4 experiments, we tested this prediction with fame judgments for names and a highly sensitive forced-choice occupation task. In Experiment 1, we demonstrated that familiar-only experiences for fame judgments are associated with above-chance performance on the semantic forced-choice occupation task. In Experiments 2 and 3, we replicated this finding and also revealed some metacognitive awareness of the availability of knowledge. In Experiment 4, we showed that graded familiarity judgments are highly correlated with the accuracy and confidence of corresponding occupation judgments. Overall, the current findings suggest that feelings of familiarity for names are not as clearly separable from semantic knowledge as the term "familiar-only" suggests. Although people may not recall a "piece" of pertinent knowledge when encountering a familiar name, this cannot be taken as evidence that no knowledge is available. These findings support the view that semantic retrieval in name recognition is better understood as a process that operates on graded evidence than as a process with discrete categorical states that only leaves an impression of familiarity when it fails.