This paper presents a novel empirical study of innovation practices of U.S. companies and their relation to productivity levels using new business micro data from the Business Research and Development and Innovation Survey (BRDIS) for the years 2008-2011. We use factor analysis to reduce a set of inputs and outputs of innovation activities into four latent unobserved innovation modes or practices. Companies are grouped according to their scores across the four factors to see that in large, small and medium companies more than one mode of innovation practices prevails. The next step in the analysis links different types of innovation practices to levels of productivity using regression analysis. The innovation modes have a statistically significant positive relation with the level of productivity. The paper demonstrates the possibility of taking into account the multidimensionality of innovation without the use of composite indicators.