Symbolic-Numeric Algebra for Polynomials (SNAP) is a prototype package that includes basic functions for computing approximate algebraic properties, such as the approximate GCD of polynomials. Our aim is to show how the unified tolerance mechanism we introduce in the package works. Using this mechanism, we can carry out approximate calculations under certified tolerances. In this article, we demonstrate the functionality of the currently released package (Version 0.2), which is downloadable from wwwmain.h.kobe-u.ac.jp/~nagasaka/research/snap/index.phtml.en. ‡ Introduction Recently, there have been many results in the area of symbolic-numeric algorithms, especially for polynomials (for example, approximate GCD for univariate polynomials [1,2,3,4] and approximate factorization for bivariate polynomials [5,6,7,8]). We think those results have practical value, which should be combined and implemented into one integrated computer algebra system such as Mathematica. In fact, Maple has such a special package called SNAP (SymbolicNumeric Algorithms for Polynomials). Currently, ours is the only such package available for Mathematica.We have been developing our SNAP package for Mathematica, which is an abbreviation for Symbolic-Numeric Algebra for Polynomials. We use algebra to mean continuous capabilities of approximate operations; for example, computing an approximate GCD between an empirical polynomial and the nearest singular polynomial computed by SNAP functions of another empirical polynomial. This continuous applicability is more important for practical computations than the number of algorithms that are already implemented, especially for average users.Our aim is to provide practical implementations of SNAP functions with a unified tolerance mechanism for polynomials like Mathematica's floating-point numbers or Kako and Sasaki's effective floating-point numbers [9]. Our idea is very simple. We only have to add new data structures representing such polynomials with tolerances and basic calculation routines and SNAP functions for the structures. At this time, only simple tolerance representations (l 2 -norm, l 1 -norm, and l ¶ -norm) for coefficient vectors of polynomials and a few SNAP functions are implemented, but the package is an ongoing project (Version 1.0 should be released in March 2007).