Previously, we proposed a two-stage model for an in vitro neural correlate of eyeblink classical conditioning involving the initial synaptic incorporation of glutamate receptor A1 (GluA1)-containing α-amino-3-hydroxy-5-methylisoxazole-4-propionic acid type receptors (AMPARs) followed by delivery of GluA4-containing AMPARs that support acquisition of conditioned responses. To test specific elements of our model for conditioning, selective knockdown of GluA4 AMPAR subunits was used using small-interfering RNAs (siRNAs). Recently, we sequenced and characterized the GluA4 subunit and its splice variants from pond turtles, Trachemys scripta elegans (tGluA4). Analysis of the relative abundance of mRNA expression by real-time RT-PCR showed that the flip/flop variants of tGluA4, tGluA4c, and a novel truncated variant tGluA4trc1 are major isoforms in the turtle brain. Here, transfection of in vitro brain stem preparations with anti-tGluA4 siRNA suppressed conditioning, tGluA4 mRNA and protein expression, and synaptic delivery of tGluA4-containing AMPARs but not tGluA1 subunits. Significantly, transfection of abducens motor neurons by nerve injections of tGluA4 flop rescue plasmid prior to anti-tGluA4 siRNA application restored conditioning and synaptic incorporation of tGluA4-containing AMPARs. In contrast, treatment with rescue plasmids for tGluA4 flip or tGluA4trc1 failed to rescue conditioning. Finally, treatment with a siRNA directed against GluA1 subunits inhibited conditioning and synaptic delivery of tGluA1-containing AMPARs and importantly, those containing tGluA4. These data strongly support our two-stage model of conditioning and our hypothesis that synaptic incorporation of tGluA4-containing AMPARs underlies the acquisition of in vitro classical conditioning. Furthermore, they suggest that tGluA4 flop may have a critical role in conditioning mechanisms compared with the other tGluA4 splice variants.