The aim of the current study was to investigate the feasibility of assessing mental load using seismocardiography by means of heart rate variability (HRV) analysis and machine learning. Twelve participants completed a mental computer task on 3 difficulty levels, on two days separated by at least a week. Electrocardiography (ECG) and seismocardiography (SCG) recordings were concurrently obtained, and a performance score based on the mental task were computed. Participants furthermore subjectively rated their mental workload (MWL) using the NASA-TLX. Cardiac cycle intervals were independently extracted from both ECG and SCG recordings and the HRV was analyzed in both the time- and frequency domain. The HRV results, subjective ratings and performance scores were statistically tested using a Two-way ANOVA with repeated measures, between days and MWL levels. Intraclass correlation coefficients (ICC) were furthermore computed to assess the agreement between the ECG and SCG based HRV. Features from the cardiac cycle segmented SCG signals were extracted and used for classification of MWL levels using machine learning. Significant differences were found for both subjective ratings and performance scores between days and MWL levels. HRV measures showed significant difference in the Peak LF measure between MWL levels. ICC values between ECG and SCG based HRV varied between poor and excellent agreement. Classification of MWL using SCG signals was unsuccessful using the included features. It can be concluded that SCG seems to be feasible for running HRV analysis due to an automatic noise removal and cardiac cycle segmentation of SCG signals being successful. However, further work is required to potentially implement successful classification of MWL using SCG.