If you are using ArcGIS you may need to convert to floating point first like (Float(Float(NIR - Red))) / (Float(Float(NIR + Red)))

Which input bands did you use and from which satellite? Did you carry out any pre-processing steps before calculating NDVI ? Which tool did you use in ArcMap and ERDAS Imagine to calculate your NDVI values? What was the formula you used?

It is pretty easy in ArcMap if you use the 'Raster Calculator' tool. With ERDAS Imagine Model Maker you can create a simple model which can be used to get the desired NDVI from NIR and Red Bands of any Satellite dataset. I have provided some screen captures of ArcMap and ERDAS. I applied the NDVI equation in 'Raster Calculator' using (Band 5 - Band 4)/( Band 5 + Band 4) for Landsat 8 bands. The result is an NDVI image with values ranging from - 1 to +1 . This is what you expect to get, right? With ERDAS model maker the attached image is self explanatory. Let me know if you need any help.

check if the denominator is equal 0, that must not happen if you want to divide two rasters, because the division with 0 is not mathematically defined

If you had applied the correct equation, (NIR-Red)/(NIR + Red), then there is no reason not to get the right NDVI values.  After pre-processing the image, please check the values of your image, specifically on your region of interest.  You can simply do this via cursor inquiry.  For instance, click on a vegetation pixel,  if there is any, and check the values of the pixels.  Do this with other land features, checking values of the RED and NIR every time.  If you think everything is okay, proceed by applying the NDVI formula to the image.  Good luck.

First of all, If you are used Landsat data, you should to chose the best image for your application, means that some type of Landsat images have no values, I do not know why? So, you should to review the processing of your data as well or you have to change it. If you are used ArcGIS for your application, you should select your bands manually via highlighting two band 3 and 4 and after that you have to press NDVI function in Image analysis tools.

If you are using ArcGIS you may need to convert to floating point first like (Float(Float(NIR - Red))) / (Float(Float(NIR + Red)))

Thank all for answering, I used Landsat data and combined 4,3,2 bands. in the ERDAS I used both model maker ( created a simple model) and  NDVI in the unsupervised window. but the result was the same. I didn't carry out any pre-processing. 

it does n't depend on the area of the region? because the area is small compared to the area of the image( it's only 7500 ha.) the region is mountainous and I think shadow is interfere 

Hi Rasoul, thanks for your reply. Can you share the Landsat Scene ID or Entity ID which you used for the NDVI. I would like to have a look myself and see if I get the same error as you do.

Hi James, thank you so much, send you required file and hope correctly got your message

Hi Rasoul, I just needed the Landsat Scene ID so that I could have a look at the particular Landsat Image from USGS EarthExplorer. But anyway thanks for sharing both the Metadata and the Shape file. I downloaded the Landsat 5 TM scene using the Entity ID provided in the Metadata: LT51660341994253ISP00. I found out that the acquisition quality of the Landsat Scene is excellent with no quality or detector issues. It does not matter if the area is small or large. I did not find any significant impact from the shadows either.

I created an NDVI Map using the same Landsat Scene which I downloaded from USGS and then clipped out your study area using the shape file which you provided. I have attached the map here. Take a look. Although there were no pre-processing operations carried out on the images, still the NDVI values are within the valid range of -1 to 1. I have two suggestions for you, if you just add Band 4 (NIR) and Band 3 (Red) into your ArcMap and simply use the formula in 'Raster Calculator' which Thomas, Eric or I have suggested, you should not have any issues in getting an NDVI Image of the whole scene. You can then mask out the study area using your shape file. If this works, then additionally I would suggest you to compare both the NDVI derived from a normal Landsat product and the Landsat Surface Reflectance product and judge which one shows the best result. Let me know if you have any futher questions.

Nice work James!  I hope it will be a smooth-sailing analysis for Rasoul after this.

thank you so much, James!

Use a conditional statement to handle divide-by-zero errors. This will happen for background pixels with a value of zero.