Objectives: To demonstrate the task-specificities of anticipatory muscle activations (AMAs) among different forward-reaching tasks and to explore the StartleReact Effect (SE) on AMAs in occurrence proportions, AMA onset latency or amplitude within these tasks in both healthy and stroke population.Methods: Ten healthy and ten stroke subjects were recruited. Participants were asked to complete the three forward-reaching tasks (reaching, reaching to grasp a ball or cup) on the left and right hand, respectively, with two different starting signals (warning-Go, 80 dB and warning-startle, 114 dB). The surface electromyography of anterior deltoid (AD), flexor carpi radialis (FCR), and extensor carpi radialis (ECR) on the moving side was recorded together with signals from bilateral sternocleidomastoid muscles (SCM), lower trapezius (LT), latissimus dorsi (LD), and tibialis anterior (TA). Proportions of valid trials, the incidence of SE, AMA incidence of each muscle, and their onset latency and amplitude were involved in analyses. The differences of these variables across different move sides (healthy, non-paretic, and paretic), normal or startle conditions, and the three tasks were explored. The ECR AMA onset was selected to further explore the SE on the incidence of AMAs.Results: Comparisons between move sides revealed a widespread AMA dysfunction in subacute stroke survivors, which was manifested as lower AMA onset incidence, changed onset latency, and smaller amplitude of AMAs in bilateral muscles. However, a significant effect of different tasks was only observed in AMA onset latency of muscle ECR (F = 3.56, p = 0.03, η2p = 0.011), but the significance disappeared in the subsequent analysis of the stroke subjects only (p > 0.05). Moreover, the following post-hoc comparison indicated significant early AMA onsets of ECR in task cup when comparing with reach (p < 0.01). For different stimuli conditions, a significance was only revealed on shortened premotor reaction time under startle for all participants (F = 60.68, p < 0.001, ηp2 = 0.056). Furthermore, stroke survivors had a significantly lower incidence of SE than healthy subjects under startle (p < 0.01). But all performed a higher incidence of ECR AMA onset (p < 0.05) than with normal signal. In addition, the incidence of ECR AMAs of both non-paretic and paretic sides could be increased significantly via startle (p ≤ 0.02).Conclusions: Healthy people have task-specific AMAs of muscle ECR when they perform forward-reaching tasks with different hand manipulations. However, this task-specific adjustment is lost in subacute stroke survivors. SE can improve the incidence of AMAs for all subjects in the forward-reaching tasks involving precision manipulations, but not change AMA onset latency and amplitude.