This dissertation investigates the application of Genetic Algorithms (GAs) to the process of implicit knowledge discovery over databases (KDD - Knowledge Discovery Database). The objective of the work has been the assessment of the Genetic Algorithms (GA) performance in the classification process of database registers. In the context of Genetic Algorithms, this classification process consists in the evolution of association rules that characterise, through its accuracy and range, a particular group of database registers. This work has encompassed four main steps: a study over the area of Knowledge Discovery Databases; the GA model definition applied to Data Mining; the implementation of the Data Mining Rule Evolver; and the case studies. The study over the KDD area included the overall process of useful knowledge discovery; the problem definition; data organisation; data pre-processing; data encoding; data improvement; data mining; and results´ interpretation. Particularly, the investigation emphasied the data mining procedure, techniques and algorithms (neural Networks, rule Induction, Statistics Models and Genetic Algorithms). A survey over the mais research projects in this area was developed from this work. The Genetic Algorithm modelling encompassed fundamentally, the definition of the chromosome representation, the fitness evaluation function and the genetic operators. Quantitative and categorical attributes must be taken into account within data mining through association rules. Quantitative attribites represent continuous variables (range of values), whereas categorical attributes are discrete variable. In the representation employed in this work, each chromosome represents a rule and each gene corresponds to a database attribute, which can be quantitative or categorical, depending on the application. The evaluation function associates a numerical value to the discovered rule, reflecting, therefore, the fitness evaluation function should drive the process towards the best association rules. The accuracy and range are performance statistics and, in some cases, their values stay nil during part of the evolutionary process. Therefore, the fitness evaluation function should reward chromosomes containing promising rules, which present accuracy and range different of zero. Ten fitness evaluation functions have been implemented. The genetic operators used in this work, crossover and mutation, seek to recombine rules´clauses in such a way to achieve rules of more accuracy and broader range when comparing the ones already sampled. Four splicing operators and two mutation operators have been experimented. The GA modeling tool implementation applied to Data Mining called Rule Evolever, evaluated the proposed model to the problem of register classification. The Rule Evolver analyses the database and extracts association rules that can better differentiate a group of registers comparing to the overall database registers. Its main features are: database attributes selection; attributes statistical information; evaluation function selection among ten implemented ones; genetic operators selection; graphical visualization of the system performance; and rules interpretation. A particular genetic operator is selected at each reproduction step, according to a previously defined rate set by the user. This rate may be kept fix or may very along the evolutionary process. The evolutionary process. The evaluation functions may also be changed (a rewarding may be included) according to the rule´s range and accuracy. The Rule Evolver implements as interface between the database and the GA, endowing the KDD process and the Data Mining phase with flexibility. In order to optimise the rules´ search process and to achieve better quality rules, some evolutionary techniques have been implemented (linear rank and elitism), and different random initialisation methods have been used as well; global average va