We compared variability, frequency composition, and temporal regularity of submaximal isometric elbow flexion force at 10, 20, 35, and 50 % of peak torque between 34 stroke subjects (5–48 days post-onset, both arms) and 24 age-matched controls (dominant arm), and related the findings in the paretic arm to motor impairment. Force variability was quantified by the coefficient of variation (CV), frequency composition by the median frequency and relative power in 0–3-, 4–6-, and 8–12-Hz bands, and regularity by the sample entropy (SampEn). The paretic elbow flexors showed significantly increased CV and relative power in 0–3-Hz band, decreased power in 4–6- and 8–12-Hz bands, and decreased SampEn compared to both the non-paretic and control elbow flexors (P ≤ 0.0002), with no differences between the latter two (P ≥ 0.012). With increasing contraction intensity, the relative power in different frequency bands was insufficiently modulated and SampEn excessively decreased in the paretic elbow flexors. Also, CV in the paretic elbow flexors was non-linearly related to the relative power in different frequency bands and SampEn across contraction intensities (rectangular hyperbolic fit, 0.21 ≤ R2 ≤ 0.55, P ≤ 0.006), whereas no force parameter correlated with arm motor impairment. These results largely extend our previous findings in the paretic knee extensors to the elbow flexors in subacute stroke, except that here force variability was increased only in the paretic elbow flexors and modulation of force regularity with increasing contraction intensity showed the opposite, decreasing pattern, which was considerably exaggerated in the paretic muscles.