It is argued from typological and perceptual evidence that the constraint against geminates (*Gem) should be split into constraints that incorporate contextual information (word position & adjacent segments). It is shown that splitting *Gem accurately captures the distribution and conspiratorial behavior of geminates in Polish.