Techniques are proposed that allow systems that use multiple oscillators to correct the effects of occasional frequency jumps, or of changes in drift rates in any of the oscillators. Existing methods of drift prediction may readily be adapted to suppress drift changes in individual oscillators. In addition, previously unreported jump detection schemes are shown to be capable of suppressing large jumps, and with very low levels of additional noise. By way of example, a system that detects jumps with single Allan-variance type measurements will limit individual jump amplitude at the output to 4·σout/√(N-1). For Flicker-FM noise, the random walk noise that this scheme generates would reach the Allan variance after about 80 000 ·τ. More complex schemes, but that still need at most 16 samples to be stored, would reduce the jump height by a factor ≫ 2, while extending the time by several orders of magnitude. This allows systems using three or more oscillators to operate in the presence of these jumps and with negligible increase in total system error. These techniques have the potential to avoid the need for costly or bulky perturbation-free oscillators in some intermediate applications.