This dissertation investigates the relationship between the acquisition orders of phonological contrasts by children in perception and production and the phonological theories that account for this relationship. Three key words can be used to characterize this relationship: gap, parallel and mismatch. It is commonly observed that young children’s ability to perceive phonological contrasts is more advanced than their ability to produce them (e.g. Smith 1973, Werker and Stager 2000). It has also been found that the order in which phonological contrasts are acquired in production recapitulates that in perception (Jusczyk et al. 1999, Pater 2004). Experiments done as part of this dissertation suggest that the parallel between perceptual and productive acquisition orders of phonemic contrasts does not always hold: 17-month-old American-English-acquiring children were able to distinguish [n] and [r] yet not [t] and [n] in a perceptual wordlearning task; while productively, the [t]-[n] contrast has been found to be acquired earlier than the [n]-[r] contrast. In other words, the orders of acquisition of phonological contrasts in perception and production can mismatch each other. Most phonological acquisition models (reviewed in this dissertation: Smith 1973, Braine 1976, Macken 1980, Boersma 1998, Smolensky 1996a, Lassettre and Donegan 1998, and Pater 2004) are able to account for the gap. The model proposed by Pater (2004) is also able to explain the parallel. When more than one phonological contrasts are involved and the order of acquisition between them is at issue, its explanation for the developmental parallel would depend on two necessary assumptions that the model did not elaborate: One, the shared MARKEDNESS constraints must be fixed in ranking; and two, the FAITHFULNESS constraints must not only be fixed in ranking, but also be homogeneous in form and function. However, under these assumptions, the model will not be able to explain the attested mismatch. This dissertation proposes to revise Pater’s model by allowing non-homogeneous faithfulness constraints for perception and production. It demonstrates how the revised model is able to account for the mismatch, explain the gap, and at the same time allow for the parallel.