An adaptive candidate approach for the fast orthogonal search (FOS) algorithm is presented for detecting signals the frequencies of which are much more closely spaced than can be resolved by the fast Fourier transform (FFT). The algorithm is shown to give solutions comparable to a global minimisation of the sum squared error, but with much greater speed. It can also surpass the Cramer-Rao lower bound on frequency resolution.Introduction: Fast orthogonal search (FOS) is a modelling technique that has been shown to be able to accurately estimate the frequency and magnitude of closely spaced signals with up to ten times the fast Fourier transform (FFT) resolution [1-3]. These results were found using a set of candidate functions with closely spaced frequencies and fitting the candidate functions one frequency at a time in order of significance. By fitting sinusoids in candidate sets of two or more frequencies and choosing the set with the lowest sum square error (SSE), it is possible to resolve candidate frequencies with an even smaller frequency separation. Simulated time series with varying signal-to-noise ratios (SNRs) are created and the adaptive FOS technique presented here is used to estimate the frequency, magnitude and phase of the closely spaced sinusoidal signals.