Mean blood pressure (MAP) can be expressed as a function of stroke volume (SV), heart rate (HR) and total peripheral resistance. The measurement of HR is easy. Continuous non-invasive measurements of MAP can be accomplished with blood pressure monitor that uses the volume-clamp method. However, the prolonged measurement with this method may yield erroneous results because the pressure in the cuff placed on a finger limits blood perfusion. Also, the mobility of a subject is restricted. SV can be estimated non-invasively for every heartbeat with impedance rheocardiography, based on changes of thorax electric impedance. This method is suitable for ambulatory measurements, but results can be disturbed by motion artefacts. In this paper we tried to predict the subsequent value of MAP basing on time series of impedance cardiography measurements (dZ/dtmax, ET, Z0, SV) and HR. For analysis, we use signals recorded in 13 young, healthy subjects, when performing three minutes handgrip test followed by two minutes of recovery. We used simple neural network model being a combination of the long short-term memory layer with the dense output layer. We divided data into sequences (data from 10 cardiac cycles and subsequent single MAP as desired output). We combined data from all recordings into one dataset. Data were normalized, assuming for each parameter, the mean value equals 0, and the standard deviation equals 1. For neural network training, we use the RMSprop optimizer, 80% of sequences were used for training and 20% for validation. The best result, the mean absolute error, in the evaluation of our model was 0.3. The efficiency of the neural network model is unsatisfactory. These results indicate that the neural network used in this study was unable to predicted MAP value from impedance rheocardiography signal.