Advantages of sequence data for reconstructing evolutionary trees include their wide scope, the large number of characters, the easier use of objective methods for building and testing trees, the use of information from mechanisms of nucleotide changes, the lower cost of obtaining information, and the predictability of finding useful characters. There are however still many problems estimating the reliability of the results of tree reconstruction. These are discussed, with examples, under the three headings of sampling error, methodological problems, and human errors. The methodological problems are the hardest to solve. They include the large number of trees, incomplete use information, inconsistency (converging to an incorrect tree), problems derived from unknown selection pressures on sequences, and trees being an inappropriate model. To overcome these problems, a good method for reconstructing trees should have the properties of being fast, efficient, consistent, robust and falsifiable. Considerable progress has been made but present methods are still best considered as 'Exploratory Data Analysis' (EDA) techniques.