Prenatal prediction in families at risk for autosomal recessive proximal spinal muscular atrophy (SMA) mainly of type I is often requested due to the high incidence and the fetal outcome of the disease. So far, only indirect genotype analysis can be performed in SMA families, since the gene has not yet been identified. We present our experience of 109 prenatal diagnoses obtained in 91 families by use of single- and multi-locus polymorphic microsatellites of the region 5q11.2-q13.3. The marker combinations and specific features of the closest microsatellites are described in detail. From 137 requests for prenatal prediction of SMA between October 1991 and August 1994, 28 families were excluded, mostly because the clinical diagnosis was uncertain or doubtful. Others had to be classified as 'SMA-variants' or showed autosomal dominant transmission of SMA. Of the 109 prenatal diagnoses performed, 29 fetuses were diagnosed to be at high risk (> 99 per cent) of developing the disease, while in seven additional pregnancies no exact prediction could be made due to a recombination event in one parental haplotype. Altogether, recombinations between closely flanking markers were observed in 14 cases. In 35 cases, the parents decided to terminate the pregnancy. Of the remaining pregnancies, 32 could be followed beyond term. All infants were reported to develop normally without signs of SMA. Two children were born with transverse reduction defects of one hand, which was most likely related to early chorionic villus sampling at 9 and 10 weeks' gestation. No further abnormalities could be detected. The limits of indirect genotype analysis and the problems of diagnostic accuracy and heterogeneity of proximal SMA are discussed.