Fetal cardiovascular monitoring is essential in assessing high-risk gestations during the third trimester. With available technologies and signal processing methods, several congenital heart diseases can be detected prenatally, and adequate treatment can be prepared in advance. Phonocardiography is one such method, with the advantage of being inexpensive and convenient compared to other commercially available methods. In this paper, an easily adaptable method is introduced to classify fetal S1 as regular or split heart sounds. The process also includes a simple heart sound segmentation, based on the Teager-Kaiser energy operator and a moving average filter. Multiple signal features were evaluated, including time domain, frequency domain, and time-frequency information. The feature-space was constructed by principal component analysis with different feature compositions and classified using k-means clustering. Classification accuracy was evaluated on a clinical recording where 1874 heart sounds had been detected. The recording was labelled by an expert with 892 regular and 762 split S1 and 220 non-decidable or noisy segments, which served as the ground truth. The highest overall accuracy achieved with individual features was 90.7%. A possible improvement upon this design using multiple classification stages based on additional features is also described.