Arterial blood gas (ABG) analysis in a newborn is a key test performed by a physician that allows to examine the acid-base balance in the body. Gasometry is also performed to assess the efficiency of the breathing process and to recommend the appropriate treatment for possible disorders. Due to the high dynamics of changes in the four measured values (pH, PaO 2 , PaCO 2 , HCO 3 ), unclear relationships between them and the unknown influence of external or individual factors the future results are extremely difficult to predict. The paper proposes a prediction algorithm which use a cyclically re-learned simple sigmoid dual-layer perceptron artificial neural network (ANN). The training based on the last few historical samples is made at every step completely from the beginning. Each time the network is used to predict only one result in the next step. The presented solution allows to forecast the gas values in the following eight hours with an average error below 1%.