Purpose: To examine the possibility that early signal-to-word form mapping capabilities are robust enough to handle substantial indexical variation in the realization of words. Method: Two groups of 7.5-month-olds were tested with the Headturn Preference Procedure. Half of the infants were exposed to words embedded in passages spoken by their mothers and tested on lists of trained and novel isolated words spoken by their fathers. The other half of the infants were yoked pairs listening to unfamiliar speakers. Results: In the test phase, infants listened longer to trained than to novel words, indicating that they successfully segmented the words from the passages. This result was not modulated by infants' familiarity with the speaker. Conclusions: Under more naturalistic listening conditions, 7.5-month-olds exhibit the ability to recognize words in the face of substantial indexical variation regardless of whether speakers are familiar. This suggests that early word representations are, at least to some extent, independent of the speaker's gender and may reflect sophisticated abstraction capabilities on the part of the infants, which would render extreme episodic models of early speech perception untenable. Additional research using similarly ecologically valid testing methods is called for to elucidate the precise nature of early word representations.