Let us explain the place of the direction, which, under the initiative of French mathematicians, was called the «data analysis» [1-4], among the problems of applied mathematics. Classic direction of the applied mathematics is connected with the methods of calculation of some characteristics of the investigated object or phenomenon from the known values of its other characteristics. In this case the object model is considered as known and the dependencies between characteristics are described by analytical expression in a form of equation or system of equations or inequalities. The problems, arising at solving of such problems, are connected with the big volume of calculations, protection from errors, accumulating in the computer due to numbers rounding etc.        Later the object analysis problems appeared, where the mathematical model is known with accuracy up to parameter values. The set of characteristics, effecting the aim characteristic, the general form of dependencies between characteristics are known, while coefficients, power index and other parameters are unknown and to determine them the protocols of observations, revealing the values of some characteristics at some values of others, are used. The series of assumptions on the values of unknown parameters are being made and these assumptions are tested on the protocols. In the result such parameters values are selected, which provide the model to define some (outlet or aim) characteristics form other (inlet) ones with required accuracy. The problems of such kind are called «model identification» problems.        At last, with the appearance of cybernetics the problems of «black box» analysis were formulated: the researcher knows the set of characteristics, containing ones influencing the aim object property, but it is unknown which of them are determining («informative») and which model describes the regularities of their effect on aim characteristic. It is necessary to select informative characteristics and to construct the model, making possible to calculate the value of aim property according to the values of other characteristics. The only one source of information for solution of such problem is the experimental data table of «incentive – reaction» with description of inlet and outlet characteristics of the observed object or multitude of objects. As we have seen above, such tables are called «object-property» type tables. Now the selection of the model and its parameters is performed by testing of different hypotheses on the data table material. Appearing range of problems forms the direction, called «data analysis» problems.        Returning to the beginning, one can note, that computing mathematics usually does not deal with the stage of creation of hypotheses of what characteristics should be involved into the object model and how the model should look like. The risk to make an error in the model and its parameters selection is beyond the area of attention and the accurate calculations with existing model make the impression of high quality of problem solution in general. The problems of model identification require the mathematician responsibility for the correct choice of model parameters. The presence of this risky step in the process of problem solution deprives the result of the strict mathematical purity aura.         The results of data analysis problems solutions shows the evident trace of a big number of heuristic or expert assumptions, concerning the selection of object characteristics, class of models, parameters of the selected model. These assumptions are presented in the language of mathematical formulas, but the nature of their appearance lies beyond the mathematics, so the main part of the process of data analysis problems solution is connected with propagation into the nature of the investigated object and is more typical for natural sciences. The situation is also aggravated by the fact that real data may have the specifics, complicating the application of strict mathematical methods. It is enough to note that the data tables are often represented by small selections in the spaces of high dimension without information about the character and degree of dependence of one characteristics upon another, with different types of measurement scales, presence of noise and empty cells. Under such conditions the data analysis problems solution methods are forced to be based both on correct mathematical procedures and on clearly heuristic tricks. It is not surprising that obtained solutions are accepted suspiciously and many solution methods don’t look strictly valid. This circumstance is reveals objectively by the fact, that at any stage of applied mathematics development there appear the real problems, requiring not existing yet well founded mathematical methods for their solution. At the same time, the importance of these problems does not permit to put their solution aside and require to accept the risky empirical hypotheses and use non-strict heuristics tricks. If the results obtained (predictions, prognoses) are confirmed by the facts, then the suspiciousness in the acceptance of used model is changed to the confidence of its adequacy to the investigated phenomenon and the attention of the mathematician is transferred to the analytical investigation of the model and computing problems, arising with its application.        The friendly and incentive remarks such as «bare heuristic», «foggy flow of literature» and so on in this case are applied by severe critics already to the attempts of the solution of problems in the boiling level of new applied problems.        OTEKS software presents the programs for solution of the most often met problems of data analysis.  REFERENCES 1. Benercri, J.P., Historie et prehistorie de l’analyse des donnees: l’analyse des correspondance. Les cahier de l’analyse des donnees, 2, 1997.   2. Data Analysis and informatics 3. Edited by E.Diday, M.Jambu, L.Lebart, J.Pages, R.Tomassone. North-Holland, Amsterdam - New York – Oxford, 1983.   3. R.Michalski, I.Bratko, M.Kubat. Mashine Learning and Data Mining. John Wiley and Sons,1989.  4. Zagoruiko N.G., Elkina V.N., Lbov G.S., Emelianov S.V. OTEKS applied sofware. «Finances and statistics», Moscow, 1986.