A sensory texture profile of cooked potatoes (Solanum tuberosum) was developed through two training periods in 1995 and 1997. Training time necessary to obtain a reliable sensory panel in assessing texture attributes was studied by the use of univariate and multivariate data analysis. The panel was interviewed and introduced to the Texture Profile Method. A list of texture attributes, ranging from non oral terms to terms evaluated after the first bite and the following mastications, was developed resulting in 14 texture attributes, Analysis of variance (ANOVA), discriminant partial least squares regression (DPLSR) and principal component analysis (PCA) was used to reduce the texture profile to 8 reliable sensory attributes covering the geometrical (reflection, mealiness, graininess), mechanical (firmness, hardness, springiness chewiness) and moistness characteristics. These variables explained 89% and 86% of the total variance in 1995 and 1997, respectively. Relevant criteria for studying panel performance are discussed. Training for more than one day did not significantly improve panel performance. Geometrical attributes as well as moistness were found easier to evaluate than the mechanical attributes.