Oversampling is the only truly effective way.
As oddson rightly said in his earlier posts, aliased frequencies get "reflected" to become other, new, frequencies that are below the Nyquist limit.
For example say you make a wave that has a fundamental of 18kHz.
Unless it is a pure sine, the second harmonic (octave) will be at 36kHz - well above Nyquist.
Lets assume that the sample frequency is 44kHz - so the Nyquist limit is 22kHz. That octave harmonic is 14kHz above the limit.
What happens is that instead of being 14kHz above the limit, it will be transformed to 14kHz below the limit - 22kHz - 14kHz = 8kHz. But 8kHz is below our original fundamental frequency, so low pass filtering is not able to get rid of the "alias".
This also demonstrates something else - the alias at 8kHz, is no longer a harmonic (or sub-harmonic) of the original 18kHz tone - which is why aliasing distortion sounds so much worse than "analogue" distortion.
SM's stock oscillators get around this by using wavetables - they do not alias because they play back short waveform samples that already have the >Nyquist frequencies removed. There are many of these short samples for each waveshape - chosen according to pitch, so that exactly the right harmonics are removed.
If your synth only mixes together (adds/subtracts) multiple oscillators, then all is well, you still have no aliasing - but if you FM or RM (multiplying oscillator outputs), then new frequencies are made that can once again be above the limit.

















