The 20th century regional and global sea level variations are estimated based on long-term tide gauge records. For this the neural network technique is utilized that connects the coastal sea level with the regional and global mean via a nonlinear empirical relationship. Two major difficulties are overcome this way: the vertical movement of tide gauges over time and the problem of what weighting function to choose for each individual tide gauge record. Neural networks are also used to fill data gaps in the tide gauge records, which is a prerequisite for our analysis technique. A suite of different gap-filling strategies is tested which provides information about stability and variance of the results. The global mean sea level for the period January 1900 to December 2006 is estimated to rise at a rate of 1.56 ± 0.25 mm/yr, which is reasonably consistent with earlier estimates, but we do not find significant acceleration. At present we consider the reconstruction of North Atlantic sea level on a basis of EOF/principal components. Components are estimated from surrounding tide gauges using neural networks while EOFs are provide by satellite altimetry