With the rapid development in seismic attribute and interpretation techniques, interpreters can be overwhelmed by the number of attributes at their disposal. Pattern recognition-driven seismic facies analysis provides a means to identify subtle variations across multiple attributes that may only be partially defined on a single attribute. Typically, interpreters intuitively choose input attributes for multiattribute facies analysis based on their experience and the geologic target of interest. However, such an approach may overlook unsuspected or subtle features hidden in the data. We therefore augment this qualitative attribute selection process with quantitative measures of candidate attributes that best differentiate features of interest. Instead of selecting a group of attributes and assuming all the selected attributes contribute equally to the facies map, we weight the interpreter-selected input attributes based on their response from the unsupervised learning algorithm and the interpreter’s knowledge. In other words, we expect the weights to represent “which attribute is ‘favored’ by an interpreter as input for unsupervised learning” from an interpretation perspective and “which attribute is ‘favored’ by the learning algorithm” from a data-driven perspective. Therefore, we claim the weights are user guided and data adaptive, as the derivation of weight for each input attribute is embedded into the learning algorithm, providing a specific measurement tailored to the selected learning algorithm, while still taking the interpreter’s knowledge into account. We develop our workflow using Barnett Shale surveys and an unsupervised self-organizing map seismic facies analysis algorithm. We found that the proposed weighting-based attribute selection method better differentiates features of interest than using equally weighted input attributes. Furthermore, the weight values provide insights into dependency among input attributes.