This research proposes a new approach, MDLR analysis, applied to stylistics, which combines a corpus linguistic approach, multi-dimensional (MD) analysis with a statistical method, linear regression (LR). MD analysis is used to study linguistic variation among different texts based on dimensions (and dimensions are underlying linguistic variables in text identified through a statistical method, factor analysis). Using MD analysis, the fiction texts from the Lancaster-Oslo/Bergen (LOB) corpus, representing the British fiction texts printed in 1961, have been characterised as neither highly involved nor informational among a wide range of written and spoken English text categories. My novel MDLR analysis approach newly combines MD analysis with linear regression to explore the relationship between a dimension identified in a study applying MD analysis to fiction texts and the stylistically relevant features of these texts. It aims to supplement the understanding of the style of the fiction population indicated by that study in terms of this dimension. Using my approach, for a given British fiction text printed in 1961, I can indicate whether this text behaves differently from the majority of the fiction texts with several identical characteristics. Moreover, two linguistically unusual texts among the fiction texts of the LOB Corpus were discovered, showing the uniqueness of style compared with the general style of British fiction in 1961. These findings supplement our understanding of the style of the British fiction texts printed in 1961, contributing to the study of the history of literature. Apart from the above application, my MDLR analysis approach was compared with an existing stylistic practice, also involving corpus linguistic approaches, to show the strengths and limitations of mine. The practice, labelled the “feature frequency hypothesis investigation (FFHI)” approach, explores a hypothesis about the frequency of a stylistically relevant linguistic feature (or features) in different texts generated through one’s reading experience. Using this approach, I preliminarily explored the evidence for a hypothesis about thought presentation in adventure fiction generated when I read the fiction texts in the LOB Corpus. The comparison shows that my MDLR approach and the FFHI approach are different in statistical complexity, exploratory or confirmatory degree and focus of contributions, but they are neither superior nor inferior to each other and are suitable for different situations. If we want to identify linguistically unusual texts in a corpus, we can choose MDLR. If we are interested in a hypothesis about a linguistic feature (or features) generated when we read the corpus, we can choose FFHI.