I did the following, now I'm getting a black screen:

void render()
{
IDirect3DSurface9*  oldBB;
device->GetRenderTarget(0, &oldBB);

device->SetRenderTarget(0, DepthSurface);
device->SetRenderTarget(1, ColorSurface);

// -- Render the scene
HRESULT hr = device->Clear(0, NULL, D3DCLEAR_TARGET | D3DCLEAR_ZBUFFER, D3DCOLOR_XRGB(0, 0, 0), 1.0f, 0);
assert(SUCCEEDED(hr));

device->BeginScene();

RenderScene();

device->SetRenderTarget(0, oldBB);
device->SetRenderTarget(1, NULL);
oldBB->Release();

d3ddev->EndScene();
d3ddev->Present(NULL, NULL, NULL, NULL);
}

If i understood d3d process correctly :

When you create the device there is a "main" backbuffer surface created for you implicitly so at last "step" you need to draw to that surface to be able to see something on screen:

void render()
{
IDirect3DSurface9*  oldBB;
device->GetRenderTarget(0, &oldBB); // grabs the last render target, and it happens to be main (backbuffer) since there is nothing set before

device->SetRenderTarget(0, DepthSurface);
device->SetRenderTarget(1, ColorSurface);

// -- Render the scene
HRESULT hr = device->Clear(0, NULL, D3DCLEAR_TARGET | D3DCLEAR_ZBUFFER, D3DCOLOR_XRGB(0, 0, 0), 1.0f, 0);
assert(SUCCEEDED(hr));

device->BeginScene();

RenderScene();

device->SetRenderTarget(0, oldBB); // backbuffer is set
device->SetRenderTarget(1, NULL);
oldBB->Release();
 
DrawSomethingThatUsesDepthAndColorTextureToBackbuffer();

d3ddev->EndScene();
d3ddev->Present(NULL, NULL, NULL, NULL); // "presents" content of the backbuffer to the screen
}

So that means I will have to return depth, color and normal with each Pixel Shader?

It is up you how many RTs you need, what do you want to store in them and when to switch to other RT...

@belfegor: There is something confuse me, I have already drawn everything when I called RenderScene();

Now what should I do in DrawSomethingThatUsesDepthAndColorTextureToBackbuffer()? Should I render the scene again?

Drawing a so-called fullscreen quad with that color render target as texture. Alternatively, switch that color and "depth" rendertarget, so that the color goes to your normal backbuffer - again.

You're now actually close to what is called deferred rendering so you might wanna read about that.

There is something confuse me, I have already drawn everything when I called RenderScene();

void Render()
{
...
device->SetRenderTarget(0, DepthSurface);
device->SetRenderTarget(1, ColorSurface);

// -- Render the scene
HRESULT hr = device->Clear(0, NULL, D3DCLEAR_TARGET | D3DCLEAR_ZBUFFER, D3DCOLOR_XRGB(0, 0, 0), 1.0f, 0);
assert(SUCCEEDED(hr));

device->BeginScene();

RenderScene();
...
}

HRESULT hr = D3DXSaveTextureToFile("depth.png", D3DXIFF_PNG, DepthTex, NULL);

device->SetRenderTarget(0, ColorSurface); // *** Switch to color surface ***

class FSQuad
{
private:
	LPDIRECT3DDEVICE9            device;
	LPDIRECT3DVERTEXBUFFER9      vBuffer;
	LPDIRECT3DINDEXBUFFER9       iBuffer;
	LPDIRECT3DVERTEXDECLARATION9 vDeclaration;

	struct FS_VERTEX
	{
		D3DXVECTOR3 position;
		D3DXVECTOR2 texcoord;
	};

public:

	FSQuad(LPDIRECT3DDEVICE9 _device)
		: device(_device), iBuffer(nullptr), vBuffer(nullptr), vDeclaration(nullptr)
	{
		device->AddRef();
	}

	~FSQuad()
	{
		if(nullptr != device)
		{
			device->Release();
			device = nullptr;
		}

		if(nullptr != iBuffer)
		{
			iBuffer->Release();
			iBuffer = nullptr;
		}

		if(nullptr != vBuffer)
		{
			vBuffer->Release();
			vBuffer = nullptr;
		}

		if(nullptr != vDeclaration)
		{
			vDeclaration->Release();
			vDeclaration = nullptr;
		}
	}

	bool init()
	{
		D3DVERTEXELEMENT9 fsVertexElements[] =
		{
			{0, 0,  D3DDECLTYPE_FLOAT3,  D3DDECLMETHOD_DEFAULT, D3DDECLUSAGE_POSITION, 0},
			{0, 12, D3DDECLTYPE_FLOAT2,  D3DDECLMETHOD_DEFAULT, D3DDECLUSAGE_TEXCOORD, 0},
			D3DDECL_END()
		};

		HRESULT hr = D3D_OK;
		hr = device->CreateVertexBuffer(sizeof(FS_VERTEX) * 4, D3DUSAGE_WRITEONLY, 0, D3DPOOL_MANAGED, &vBuffer, nullptr);
		if(FAILED(hr))
		{
			MessageBox(0, "Failed to create FS VB!", 0, 0);
			return false;
		}

		hr = device->CreateIndexBuffer(sizeof(USHORT) * 6, D3DUSAGE_WRITEONLY, D3DFMT_INDEX16, D3DPOOL_MANAGED, &iBuffer, nullptr);
		if(FAILED(hr))
		{
			MessageBox(0, "Failed to create FS IB!", 0, 0);
			return false;
		}

		hr = device->CreateVertexDeclaration(fsVertexElements, &vDeclaration);
		if(FAILED(hr))
		{
			MessageBox(0, "Failed to create FS VD!", 0, 0);
			return false;
		}

		float c = 1.0f;
		float z = 1.0f;
		FS_VERTEX* pVertices = nullptr;
		vBuffer->Lock(0, 0, (void**)&pVertices, 0);
		pVertices[0].position = D3DXVECTOR3( -c,  c,  z); pVertices[0].texcoord = D3DXVECTOR2(0.0f, 0.0f);
		pVertices[1].position = D3DXVECTOR3(  c,  c,  z); pVertices[1].texcoord = D3DXVECTOR2(1.0f, 0.0f);
		pVertices[2].position = D3DXVECTOR3(  c, -c,  z); pVertices[2].texcoord = D3DXVECTOR2(1.0f, 1.0f);
		pVertices[3].position = D3DXVECTOR3( -c, -c,  z); pVertices[3].texcoord = D3DXVECTOR2(0.0f, 1.0f);
		vBuffer->Unlock();

		USHORT* pIndices = nullptr;
		iBuffer->Lock(0, 0, (void**)&pIndices, 0);
		pIndices[0]       = 0;
		pIndices[1]       = 1;
		pIndices[2]       = 2;
		pIndices[3]       = 0;
		pIndices[4]       = 2;
		pIndices[5]       = 3;
		iBuffer->Unlock();

		return true;
	}

	void draw()
	{
		device->SetStreamSource(0, vBuffer, 0, sizeof(FS_VERTEX));
		device->SetIndices(iBuffer);
		device->SetVertexDeclaration(vDeclaration);
		device->DrawIndexedPrimitive(D3DPT_TRIANGLELIST, 0, 0, 4, 0, 2);
	}
};

float4 ScreenParams; // xy width/height of screen

struct VERTEX
{
	float3 Position  : POSITION;
	float2 TexCoord0 : TEXCOORD0;
};

struct VS_OUTPUT
{
	float4 Position  : POSITION;
	float2 TexCoord0 : TEXCOORD0;
};

struct PS_OUTPUT
{
    float4 Color : COLOR0;
};

void VertexProgram(in VERTEX IN, out VS_OUTPUT OUT)
{
	float2 texelOffset = float2(-1.0f / ScreenParams.x, 1.0f / ScreenParams.y);
	float2 result      = IN.Position.xy;
    result.xy         += texelOffset;
	OUT.Position.x     = result.x;
	OUT.Position.y     = result.y;
	OUT.Position.z     = 1.0f;
	OUT.Position.w     = 1.0f;
	OUT.TexCoord0      = float2(0.5f, -0.5f) * IN.Position.xy + 0.5f.xx;
}

void PixelProgram(in VS_OUTPUT IN, out PS_OUTPUT OUT)
{
    ...
    OUT.Color = ...;
}

I'm still getting a window screenshot when I execute this line:

HRESULT hr = D3DXSaveTextureToFile("depth.dds", D3DXIFF_DDS, DepthTex, NULL);

It seems that the render target "DepthTex" is getting color instead of depth map.

1. Where do you call it?

2. Does the camera look at something "useful" at the time when you take the screenshot?

3. What is stored in DepthTex?

1. I'm calling it after device->Present()

2. The camera is looking at far and close mountains, so I think that's a good scene

3. I'm getting a screenshot of the scene instead of the depth texture

I'm sure that D3D9 is writing color to the first render target instead of depth map.