Dalvik is two things -- a class library (analogous to a J2SE or J2ME implementation -- and Dalvik's doesn't match either), which is where your statement holds true, and the bytecode interpreter, which differs in less clear-cut ways.

The bytecode interpreter differs from many Java implementations because it _is_ a pure interpreter, not a JIT compiler. Also, and most significantly IMO, it's a register-based (instead of stack-based) VM. For those not familiar with this architecture debate, think of RISC/CISC, and don't ask "which is better" unless you like flamewars.