We investigate the viability of sub-millimeter wavelength oscillating deviations from the Newtonian potential at both the theoretical and the experimental/observational level. At the theoretical level such deviations are generic predictions in a wide range of extensions of General Relativity (GR) including f (R) theories, massive Brans-Dicke (BD)-scalar tensor theories, compactified extra dimension models and nonlocal extensions of GR. However, the range of parameters associated with such oscillating deviations is usually connected with instabilities present at the perturbative level. An important exception emerges in nonlocal gravity theories where oscillating deviations from Newtonian potential occur naturally on sub-millimeter scales without any instabilities. As an example of a model with unstable Newtonian oscillations we review an f (R) expansion around General Relativity of the form f (R) = R + 1 6m 2 R 2 with m 2 < 0 pointing out possible stabilization mechanisms. As an example of a model with stable Newtonian oscillations we discuss nonlocal gravity theories. If such oscillations are realized in Nature on sub-millimeter scales, a signature is expected in torsion balance experiments testing the validity of Newton's law. We search for such a signature in the torsion balance data of the Washington experiment [1] (combined torque residuals of experiments I, II, III) testing Newton's law at sub-millimeter scales. We show that an oscillating residual ansatz with spatial wavelength λ 0.1mm provides a better fit to the data compared to the residual Newtonian constant ansatz by ∆χ 2 = −15. Similar improved fits however, also occur in about 10% of Monte Carlo realization of Newtonian data on similar or larger scales. Thus, the significance level of this improved fit is at a level of not more than 2σ. The energy scale corresponding to this best fit wavelength is identical to the dark energy length scale λ de ≡ 4 c/ρ de ≈ 0.1mm.