The term curve fitting is more general than regression. Your approach to curve fitting depends on your goal. In some circumstances, your goal is simple. You don't care about models, and don't expect best-fit values that you can interpret. Instead, you just want to draw a smooth curve to make a graph look attractive, or to use as a standard curve.   

Prism provides two approaches for fitting a curve without selecting a model. A cubic spline curve goes through every data point, bending and twisting as needed. A lowess curve follows the trend of the data. Lowess curves can be helpful when the data progresses monotonically, but are less helpful when there are peaks or valleys. Prism lets you choose between fine, medium and course lowess curves. The fine curve reveals the fine structure of the data, but tends to wiggle a lot. The coarse curve shows only the general trend, but obscures the detail.

To create a lowess or spline curve, click the Analyze button and choose Fit spline/lowess from the list of curves and regressions to bring up the dialog.

Prism generates lowess curves using an algorithm adapted from Graphical Methods for Data Analysis, John Chambers et. al., Wadsworth and Brooks, 1983. Don't select a lowess curve unless you have well over twenty data points. Prism generates the curve as a series of line segments. Enter the number of segments you want, and check the option box if you need to see the XY coordinates of each point. To use the lowess, point-to-point, or spline curve as a standard curve, see Introduction to standard curves.

Prism can also create a point-to-point "curve" -- a series of line segments connecting all your data. Don't create a point-to-point curve just so you can connect points with a line on the graph. You can do that by checking an option on the Symbols & Lines dialog from the Graphs section of your project. Only select the point-to-point analysis if you want to use the point-to-point line as a standard curve or to calculate area under the curve.

Spline curves can wiggle too much. Lowess curves can be too jagged. To get a smoother curve, consider using nonlinear regression and pick a model empirically. You don't have to pick a sensible model, and don't have to interpret the best-fit values. Use nonlinear regression to create a smooth curve, not as a method to analyze data. Polynomial models are often used for this purpose. See Polynomial regression