Abstract. We examine the relation between predicative recurrence and strictly finitistic tenets in the philosophy of mathematics, primarily by focusing on the r61e of numerical notations in computing. After an overview of Wittgenstein's ideas on the "surveyability" of notations, we analyze a subtle form of circularity in the usual justification of the primitive recursive definition of exponentiation (Isles 1992), and suggest connections with recent works on predicative recurrence (Leivant 1993b, Bellantoni & Cook 1993.A long-standing thread in the foundations of mathematics questions the ontology of numeric terms and the absoluteness of the informal notion of finiteness. This paper is a preliminary exploration of some of the ideas (and puzzles) underlying these "strictly finitistic" approaches, especially in the light of recent works that relate predicative formalisms to computational complexity (Nelson 1986, Buss 1986, Leivant 1993a. One would expect investigations in these areas to lead to a unified perspective, thereby contributing to an assessment of the Karp-Cook Thesis on identifying feasible computability with poly-time (see (Davis 1982); this identification goes back to Edmonds 1965).Large finite collections, and the natural numbers that count them, are the main source of perplexity for strict finitists: their acceptance makes the ontological status of constructions requiring an arbitrarily large but finite number of steps indistinguishable from that of the objects of platonistic mathematics. We shall * Address: Dipartimento di Scienze dell'Informazione, via Comelico 39/41, 1-20135 Milano, Italy. E-mail: felice~imiucca, csi.unimi.it. This is an expanded version of my talk for the Logic and Computational Complexity Meeting, and is part of a larger project analyzing the confluence of mathematical results and philosophical arguments i,i the analysis of feasible computations that will be described in a future joint paper with Daniel Leivant, to whom I am greatly indebted for stimulating my interest in this field and for prompting me to write this preliminary account. It will be apparent from the text that his views, and also those of David Isles, have been quite influential on the present paper (of course they are not at all responsible for any of my mistakes). I am grateful also to Stephen Bellantoni, Paolo Boldi, Gabriele Lolli, Diego Marconi, Piergiorgio Odifreddi and Nicoletta Sabadini for criticisms and suggestions, and to Roberta Mari for playing both as Proponent and Opponent in the design of the games described in the last section. Finally, I would like to thank my friend, Miss Claudia Bonino, for betting long ago that this paper would eventually have been written. The research was supported by CNR Cooperation Project "Linguaggi Applicativi e Dimostrazioni Costruttive" and MURST 40%.